Repository Public
1---2title: Agents3url: https://platform.claude.com/docs/en/api/beta/agents4---56# Agents78## Create Agent910**POST** `/v1/agents`1112Create Agent1314### Headers1516- `"anthropic-beta": optional array of AnthropicBeta`1718 Optional header to specify the beta version(s) you want to use.1920 - `string`2122 - `"message-batches-2024-09-24" or "prompt-caching-2024-07-31" or "computer-use-2024-10-22" or 43 more`2324 - `"message-batches-2024-09-24"`2526 - `"prompt-caching-2024-07-31"`2728 - `"computer-use-2024-10-22"`2930 - `"computer-use-2025-01-24"`3132 - `"pdfs-2024-09-25"`3334 - `"token-counting-2024-11-01"`3536 - `"token-efficient-tools-2025-02-19"`3738 - `"output-128k-2025-02-19"`3940 - `"files-api-2025-04-14"`4142 - `"mcp-client-2025-04-04"`4344 - `"mcp-client-2025-11-20"`4546 - `"dev-full-thinking-2025-05-14"`4748 - `"interleaved-thinking-2025-05-14"`4950 - `"code-execution-2025-05-22"`5152 - `"extended-cache-ttl-2025-04-11"`5354 - `"context-1m-2025-08-07"`5556 - `"context-management-2025-06-27"`5758 - `"model-context-window-exceeded-2025-08-26"`5960 - `"skills-2025-10-02"`6162 - `"fast-mode-2026-02-01"`6364 - `"output-300k-2026-03-24"`6566 - `"user-profiles-2026-03-24"`6768 - `"user-profiles-2026-08-18"`6970 - `"user-profiles-2026-09-04"`7172 - `"advisor-tool-2026-03-01"`7374 - `"managed-agents-2026-04-01"`7576 - `"cache-diagnosis-2026-04-07"`7778 - `"dreaming-2026-04-21"`7980 - `"thinking-token-count-2026-05-13"`8182 - `"server-side-fallback-2026-06-01"`8384 - `"server-side-fallback-2026-07-01"`8586 - `"fallback-credit-2026-06-01"`8788 - `"fallback-credit-2026-07-01"`8990 - `"agent-memory-2026-07-22"`9192 - `"mid-conversation-tool-changes-2026-07-01"`9394 - `"compact-2026-01-12"`9596 - `"computer-use-2025-11-24"`9798 - `"mcp-tunnels-2026-06-22"`99100 - `"structured-outputs-2025-11-13"`101102 - `"task-budgets-2026-03-13"`103104 - `"thinking-display-updates-2026-08-18"`105106 - `"ce-user-management-2026-07-13"`107108 - `"mid-conversation-output-config-2026-07-01"`109110 - `"thinking-binding-controls-2026-08-01"`111112 - `"mid-conversation-system-clear-at-2026-08-21"`113114 - `"compact-2026-09-04"`115116- `"anthropic-workspace-id": optional string`117118### Body parameters119120- `model: BetaManagedAgentsModel or BetaManagedAgentsModelConfigParams`121122 Model identifier. Accepts the [model string](https://platform.claude.com/docs/en/about-claude/models/overview#latest-models-comparison), e.g. `claude-opus-5`, or a `model_config` object for additional configuration control123124 - `BetaManagedAgentsModel = "claude-fable-5-1" or "claude-sonnet-5" or "claude-fable-5" or 11 more or string`125126 The model that will power your agent.127128 See [models](https://docs.anthropic.com/en/docs/models-overview) for additional details and options.129130 - `"claude-fable-5-1" or "claude-sonnet-5" or "claude-fable-5" or 11 more`131132 The model that will power your agent.133134 See [models](https://docs.anthropic.com/en/docs/models-overview) for additional details and options.135136 - `"claude-fable-5-1"`137138 Frontier intelligence for ambitious tasks across coding, scientific discovery, and enterprise workflows139140 - `"claude-sonnet-5"`141142 High-performance model for coding and agents143144 - `"claude-fable-5"`145146 Next generation of intelligence for the hardest knowledge work and coding problems147148 - `"claude-opus-5"`149150 Powerful intelligence for long-running agents and coding151152 - `"claude-opus-4-8"`153154 Powerful intelligence for long-running agents and coding155156 - `"claude-opus-4-7"`157158 Powerful intelligence for long-running agents and coding159160 - `"claude-opus-4-6"`161162 Powerful intelligence for long-running agents and coding163164 - `"claude-sonnet-4-6"`165166 Best combination of speed and intelligence167168 - `"claude-haiku-4-5"`169170 Fastest model with near-frontier intelligence171172 - `"claude-haiku-4-5-20251001"`173174 Fastest model with near-frontier intelligence175176 - `"claude-opus-4-5"`177178 Powerful intelligence for long-running agents and coding179180 - `"claude-opus-4-5-20251101"`181182 Powerful intelligence for long-running agents and coding183184 - `"claude-sonnet-4-5"`185186 High-performance model for agents and coding187188 - `"claude-sonnet-4-5-20250929"`189190 High-performance model for agents and coding191192 - `string`193194 - `BetaManagedAgentsModelConfigParams object`195196 An object that defines additional configuration control over model use197198 - `id: BetaManagedAgentsModel`199200 The model that will power your agent.201202 See [models](https://docs.anthropic.com/en/docs/models-overview) for additional details and options.203204 - `effort: optional "low" or "medium" or "high" or 2 more or BetaManagedAgentsEffortLow or BetaManagedAgentsEffortMedium or 3 more or null`205206 How hard Claude works on each inference call. Accepts a bare level string (`"high"`) or `{"type": "high"}`. On create, omitting it resolves the per-model default; on update, omitting it leaves the stored value unchanged.207208 - `BetaManagedAgentsEffortLevel = "low" or "medium" or "high" or 2 more`209210 How hard Claude works on each turn. Higher levels favor reasoning depth over latency. Not all models accept every level; invalid combinations are rejected at create time.211212 - `"low"`213214 - `"medium"`215216 - `"high"`217218 - `"xhigh"`219220 - `"max"`221222 - `BetaManagedAgentsEffortLow object`223224 Low effort. Favors latency over reasoning depth.225226 - `type: "low"`227228 - `BetaManagedAgentsEffortMedium object`229230 Medium effort. Balances latency and reasoning depth.231232 - `type: "medium"`233234 - `BetaManagedAgentsEffortHigh object`235236 High effort. Favors reasoning depth.237238 - `type: "high"`239240 - `BetaManagedAgentsEffortXhigh object`241242 Extra-high effort. Not all models accept this level.243244 - `type: "xhigh"`245246 - `BetaManagedAgentsEffortMax object`247248 Maximum effort. Favors reasoning depth over latency.249250 - `type: "max"`251252 - `inference_geo: optional string or null`253254 Geographic region for model inference. When unset, requests fall through to the workspace's default_inference_geo. On update, `model` is whole-object replacement — omitting inference_geo clears it.255256 - `speed: optional "standard" or "fast" or null`257258 Inference speed mode. `fast` provides significantly faster output token generation at premium pricing. Not all models support `fast`; invalid combinations are rejected at create time.259260 - `"standard"`261262 - `"fast"`263264- `name: string`265266 Human-readable name for the agent.267268 minLength: 1, maxLength: 256269270- `description: optional string or null`271272 Description of what the agent does.273274 maxLength: 2048275276- `mcp_servers: optional array of BetaManagedAgentsURLMCPServerParams`277278 MCP servers this agent connects to. Maximum 20. Names must be unique within the array. Every server must be referenced by an `mcp_toolset` in `tools`; unreferenced servers are rejected. See the [MCP connector guide](https://platform.claude.com/docs/en/managed-agents/mcp-connector).279280 - `type: "url"`281282 - `name: string`283284 Unique name for this server, referenced by mcp_toolset configurations. 1-255 characters.285286 minLength: 1, maxLength: 255287288 - `url: string`289290 Endpoint URL for the MCP server.291292 maxLength: 2048293294- `metadata: optional map[string]`295296 Arbitrary key-value metadata. Maximum 16 pairs, keys up to 64 chars, values up to 512 chars.297298- `multiagent: optional BetaManagedAgentsMultiagentParams or null`299300 A coordinator topology: the session's primary thread orchestrates work by spawning session threads, each running an agent drawn from the `agents` roster.301302 - `type: "coordinator"`303304 - `agents: array of BetaManagedAgentsMultiagentRosterEntryParams`305306 Agents the coordinator may spawn as session threads. 1–20 entries. Each entry is an agent ID string, a versioned `{"type":"agent","id","version"}` reference, or `{"type":"self"}` to allow recursive self-invocation. Entries must reference distinct agents (after resolving `self` and string forms); at most one `self`. Referenced agents must exist, must not be archived, and must not themselves have `multiagent` set (depth limit 1).307308 - `string`309310 - `BetaManagedAgentsAgentParams object`311312 Specification for an Agent. Provide a specific `version` or use the short-form `agent="agent_id"` for the most recent version313314 - `type: "agent"`315316 - `id: string`317318 The `agent` ID.319320 minLength: 1, maxLength: 128321322 - `version: optional number`323324 The specific `agent` version to use. Omit to use the latest version. Must be at least 1 if specified.325326 format: int32327328 - `BetaManagedAgentsMultiagentSelfParams object`329330 Sentinel roster entry meaning "the agent that owns this configuration". Resolved server-side to a concrete agent reference.331332 - `type: "self"`333334 - `BetaManagedAgentsAdvisorParams object`335336 Platform advisor roster entry: a model the session's primary thread may consult mid-turn. At most one per roster; the entry occupies the roster name `anthropic.advisor`.337338 - `type: "advisor"`339340 - `model: string`341342 A Claude model id. The model must be permitted as an advisor for this agent's model — see the sessions/threads/advisor spec.343344 minLength: 1, maxLength: 256345346- `skills: optional array of BetaManagedAgentsSkillParams`347348 Skills available to the agent.349350 - `BetaManagedAgentsAnthropicSkillParams object`351352 An Anthropic-managed skill.353354 - `type: "anthropic"`355356 - `skill_id: string`357358 Identifier of the Anthropic skill (e.g., "xlsx").359360 minLength: 1, maxLength: 64361362 - `version: optional string or null`363364 Version to pin. Defaults to latest if omitted.365366 minLength: 1, maxLength: 64367368 - `BetaManagedAgentsCustomSkillParams object`369370 A user-created custom skill.371372 - `type: "custom"`373374 - `skill_id: string`375376 Tagged ID of the custom skill (e.g., "skill_01XJ5...").377378 minLength: 1, maxLength: 64379380 - `version: optional string or null`381382 Version to pin. Defaults to latest if omitted.383384 minLength: 1, maxLength: 64385386- `system: optional string or null`387388 System prompt for the agent.389390 maxLength: 100000391392- `tools: optional array of BetaManagedAgentsAgentToolset20260401Params or BetaManagedAgentsMCPToolsetParams or BetaManagedAgentsCustomToolParams`393394 Tool configurations available to the agent. Maximum of 128 tools across all toolsets allowed.395396 - `BetaManagedAgentsAgentToolset20260401Params object`397398 Configuration for built-in agent tools. Use this to enable or disable groups of tools available to the agent.399400 - `type: "agent_toolset_20260401"`401402 - `configs: optional array of BetaManagedAgentsAgentToolConfigParams`403404 Per-tool configuration overrides.405406 - `BetaManagedAgentsBashToolConfigParams object`407408 Configuration override for the bash tool.409410 - `type: optional "bash"`411412 - `name: "bash"`413414 Must be "bash".415416 - `enabled: optional boolean or null`417418 Whether this tool is enabled and available to Claude. Overrides the default_config setting.419420 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`421422 Permission policy for tool execution.423424 - `BetaManagedAgentsAlwaysAllowPolicy object`425426 Tool calls are automatically approved without user confirmation.427428 - `type: "always_allow"`429430 - `BetaManagedAgentsAlwaysAskPolicy object`431432 Tool calls require user confirmation before execution.433434 - `type: "always_ask"`435436 - `BetaManagedAgentsAutoPolicy object`437438 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.439440 - `type: "auto"`441442 - `BetaManagedAgentsEditToolConfigParams object`443444 Configuration override for the edit tool.445446 - `type: optional "edit"`447448 - `name: "edit"`449450 Must be "edit".451452 - `enabled: optional boolean or null`453454 Whether this tool is enabled and available to Claude. Overrides the default_config setting.455456 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`457458 Permission policy for tool execution.459460 - `BetaManagedAgentsAlwaysAllowPolicy object`461462 Tool calls are automatically approved without user confirmation.463464 - `BetaManagedAgentsAlwaysAskPolicy object`465466 Tool calls require user confirmation before execution.467468 - `BetaManagedAgentsAutoPolicy object`469470 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.471472 - `BetaManagedAgentsReadToolConfigParams object`473474 Configuration override for the read tool.475476 - `type: optional "read"`477478 - `name: "read"`479480 Must be "read".481482 - `enabled: optional boolean or null`483484 Whether this tool is enabled and available to Claude. Overrides the default_config setting.485486 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`487488 Permission policy for tool execution.489490 - `BetaManagedAgentsAlwaysAllowPolicy object`491492 Tool calls are automatically approved without user confirmation.493494 - `BetaManagedAgentsAlwaysAskPolicy object`495496 Tool calls require user confirmation before execution.497498 - `BetaManagedAgentsAutoPolicy object`499500 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.501502 - `BetaManagedAgentsWriteToolConfigParams object`503504 Configuration override for the write tool.505506 - `type: optional "write"`507508 - `name: "write"`509510 Must be "write".511512 - `enabled: optional boolean or null`513514 Whether this tool is enabled and available to Claude. Overrides the default_config setting.515516 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`517518 Permission policy for tool execution.519520 - `BetaManagedAgentsAlwaysAllowPolicy object`521522 Tool calls are automatically approved without user confirmation.523524 - `BetaManagedAgentsAlwaysAskPolicy object`525526 Tool calls require user confirmation before execution.527528 - `BetaManagedAgentsAutoPolicy object`529530 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.531532 - `BetaManagedAgentsGlobToolConfigParams object`533534 Configuration override for the glob tool.535536 - `type: optional "glob"`537538 - `name: "glob"`539540 Must be "glob".541542 - `enabled: optional boolean or null`543544 Whether this tool is enabled and available to Claude. Overrides the default_config setting.545546 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`547548 Permission policy for tool execution.549550 - `BetaManagedAgentsAlwaysAllowPolicy object`551552 Tool calls are automatically approved without user confirmation.553554 - `BetaManagedAgentsAlwaysAskPolicy object`555556 Tool calls require user confirmation before execution.557558 - `BetaManagedAgentsAutoPolicy object`559560 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.561562 - `BetaManagedAgentsGrepToolConfigParams object`563564 Configuration override for the grep tool.565566 - `type: optional "grep"`567568 - `name: "grep"`569570 Must be "grep".571572 - `enabled: optional boolean or null`573574 Whether this tool is enabled and available to Claude. Overrides the default_config setting.575576 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`577578 Permission policy for tool execution.579580 - `BetaManagedAgentsAlwaysAllowPolicy object`581582 Tool calls are automatically approved without user confirmation.583584 - `BetaManagedAgentsAlwaysAskPolicy object`585586 Tool calls require user confirmation before execution.587588 - `BetaManagedAgentsAutoPolicy object`589590 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.591592 - `BetaManagedAgentsWebFetchToolConfigParams object`593594 Configuration override for the web_fetch tool.595596 - `type: optional "web_fetch"`597598 - `name: "web_fetch"`599600 Must be "web_fetch".601602 - `allowed_domains: optional array of string`603604 Only fetch URLs whose host is one of these domains or a subdomain of one. Each entry is a plain hostname like "docs.example.com" (no scheme, port, or path). At most 64 entries; an empty list is rejected (omit the field instead). Cannot be combined with blocked_domains.605606 - `blocked_domains: optional array of string`607608 Never fetch URLs whose host is one of these domains or a subdomain of one. Each entry is a plain hostname like "ads.example.com" (no scheme, port, or path). At most 64 entries; an empty list is rejected (omit the field instead). Cannot be combined with allowed_domains.609610 - `enabled: optional boolean or null`611612 Whether this tool is enabled and available to Claude. Overrides the default_config setting.613614 - `max_content_tokens: optional number or null`615616 Maximum number of tokens of fetched text content to include in context per call. Does not apply to binary content such as PDFs.617618 format: int32619620 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`621622 Permission policy for tool execution.623624 - `BetaManagedAgentsAlwaysAllowPolicy object`625626 Tool calls are automatically approved without user confirmation.627628 - `BetaManagedAgentsAlwaysAskPolicy object`629630 Tool calls require user confirmation before execution.631632 - `BetaManagedAgentsAutoPolicy object`633634 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.635636 - `BetaManagedAgentsWebSearchToolConfigParams object`637638 Configuration override for the web_search tool.639640 - `type: optional "web_search"`641642 - `name: "web_search"`643644 Must be "web_search".645646 - `allowed_domains: optional array of string`647648 Only return search results whose host is one of these domains or a subdomain of one. Each entry is a plain hostname like "docs.example.com" (no scheme or port; an optional path suffix is accepted). At most 64 entries; an empty list is rejected (omit the field instead). Cannot be combined with blocked_domains.649650 - `blocked_domains: optional array of string`651652 Never return search results whose host is one of these domains or a subdomain of one. Each entry is a plain hostname like "ads.example.com" (no scheme or port; an optional path suffix is accepted). At most 64 entries; an empty list is rejected (omit the field instead). Cannot be combined with allowed_domains.653654 - `enabled: optional boolean or null`655656 Whether this tool is enabled and available to Claude. Overrides the default_config setting.657658 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`659660 Permission policy for tool execution.661662 - `BetaManagedAgentsAlwaysAllowPolicy object`663664 Tool calls are automatically approved without user confirmation.665666 - `BetaManagedAgentsAlwaysAskPolicy object`667668 Tool calls require user confirmation before execution.669670 - `BetaManagedAgentsAutoPolicy object`671672 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.673674 - `user_location: optional BetaManagedAgentsUserLocation or null`675676 Approximate user location for search result localization.677678 - `type: "approximate"`679680 Location precision. Only "approximate" is supported.681682 - `city: optional string or null`683684 City name.685686 minLength: 1, maxLength: 255687688 - `country: optional string or null`689690 Two-letter ISO 3166-1 country code, uppercase.691692 - `region: optional string or null`693694 Region or state name.695696 minLength: 1, maxLength: 255697698 - `timezone: optional string or null`699700 IANA timezone identifier, e.g. "America/Los_Angeles".701702 minLength: 1, maxLength: 255703704 - `default_config: optional BetaManagedAgentsAgentToolsetDefaultConfigParams or null`705706 Default configuration for all tools in a toolset.707708 - `enabled: optional boolean or null`709710 Whether tools are enabled and available to Claude by default. Defaults to true if not specified.711712 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`713714 Permission policy for tool execution.715716 - `BetaManagedAgentsAlwaysAllowPolicy object`717718 Tool calls are automatically approved without user confirmation.719720 - `BetaManagedAgentsAlwaysAskPolicy object`721722 Tool calls require user confirmation before execution.723724 - `BetaManagedAgentsAutoPolicy object`725726 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.727728 - `BetaManagedAgentsMCPToolsetParams object`729730 Configuration for tools from an MCP server defined in `mcp_servers`.731732 - `type: "mcp_toolset"`733734 - `mcp_server_name: string`735736 Name of the MCP server. Must match a server name from the mcp_servers array. 1-255 characters.737738 minLength: 1, maxLength: 255739740 - `configs: optional array of BetaManagedAgentsMCPToolConfigParams`741742 Per-tool configuration overrides.743744 - `name: string`745746 Name of the MCP tool to configure. 1-128 characters.747748 minLength: 1, maxLength: 128749750 - `enabled: optional boolean or null`751752 Whether this tool is enabled. Overrides the `default_config` setting.753754 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`755756 Permission policy for tool execution.757758 - `BetaManagedAgentsAlwaysAllowPolicy object`759760 Tool calls are automatically approved without user confirmation.761762 - `BetaManagedAgentsAlwaysAskPolicy object`763764 Tool calls require user confirmation before execution.765766 - `BetaManagedAgentsAutoPolicy object`767768 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.769770 - `default_config: optional BetaManagedAgentsMCPToolsetDefaultConfigParams or null`771772 Default configuration for all tools from an MCP server.773774 - `enabled: optional boolean or null`775776 Whether tools are enabled by default. Defaults to true if not specified.777778 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`779780 Permission policy for tool execution.781782 - `BetaManagedAgentsAlwaysAllowPolicy object`783784 Tool calls are automatically approved without user confirmation.785786 - `BetaManagedAgentsAlwaysAskPolicy object`787788 Tool calls require user confirmation before execution.789790 - `BetaManagedAgentsAutoPolicy object`791792 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.793794 - `BetaManagedAgentsCustomToolParams object`795796 A custom tool that is executed by the API client rather than the agent. When the agent calls this tool, an `agent.custom_tool_use` event is emitted and the session goes idle, waiting for the client to provide the result via a `user.custom_tool_result` event.797798 - `type: "custom"`799800 - `description: string`801802 Description of what the tool does, shown to the agent to help it decide when to use the tool.803804 minLength: 1805806 - `input_schema: BetaManagedAgentsCustomToolInputSchema`807808 JSON Schema for custom tool input parameters.809810 - `type: "object"`811812 - `properties: optional map[unknown] or null`813814 - `required: optional array of string or null`815816 - `name: string`817818 Unique name for the tool. 1-128 characters; letters, digits, underscores, and hyphens.819820 minLength: 1, maxLength: 128821822### Returns823824- `BetaManagedAgentsAgent object`825826 A Managed Agents `agent`.827828 - `type: "agent"`829830 - `id: string`831832 - `archived_at: string or null`833834 A timestamp in RFC 3339 format835836 format: date-time837838 - `created_at: string`839840 A timestamp in RFC 3339 format841842 format: date-time843844 - `description: string or null`845846 - `mcp_servers: array of BetaManagedAgentsMCPServerURLDefinition`847848 - `type: "url"`849850 - `name: string`851852 - `url: string`853854 - `metadata: map[string]`855856 - `model: BetaManagedAgentsModelConfig`857858 Model identifier and configuration.859860 - `id: BetaManagedAgentsModel`861862 The model that will power your agent.863864 See [models](https://docs.anthropic.com/en/docs/models-overview) for additional details and options.865866 - `"claude-fable-5-1" or "claude-sonnet-5" or "claude-fable-5" or 11 more`867868 The model that will power your agent.869870 See [models](https://docs.anthropic.com/en/docs/models-overview) for additional details and options.871872 - `"claude-fable-5-1"`873874 Frontier intelligence for ambitious tasks across coding, scientific discovery, and enterprise workflows875876 - `"claude-sonnet-5"`877878 High-performance model for coding and agents879880 - `"claude-fable-5"`881882 Next generation of intelligence for the hardest knowledge work and coding problems883884 - `"claude-opus-5"`885886 Powerful intelligence for long-running agents and coding887888 - `"claude-opus-4-8"`889890 Powerful intelligence for long-running agents and coding891892 - `"claude-opus-4-7"`893894 Powerful intelligence for long-running agents and coding895896 - `"claude-opus-4-6"`897898 Powerful intelligence for long-running agents and coding899900 - `"claude-sonnet-4-6"`901902 Best combination of speed and intelligence903904 - `"claude-haiku-4-5"`905906 Fastest model with near-frontier intelligence907908 - `"claude-haiku-4-5-20251001"`909910 Fastest model with near-frontier intelligence911912 - `"claude-opus-4-5"`913914 Powerful intelligence for long-running agents and coding915916 - `"claude-opus-4-5-20251101"`917918 Powerful intelligence for long-running agents and coding919920 - `"claude-sonnet-4-5"`921922 High-performance model for agents and coding923924 - `"claude-sonnet-4-5-20250929"`925926 High-performance model for agents and coding927928 - `string`929930 - `effort: optional BetaManagedAgentsEffortLow or BetaManagedAgentsEffortMedium or BetaManagedAgentsEffortHigh or 2 more`931932 How hard Claude works on each turn. Sets `output_config.effort` on every Messages call the session makes.933934 - `BetaManagedAgentsEffortLow object`935936 Low effort. Favors latency over reasoning depth.937938 - `type: "low"`939940 - `BetaManagedAgentsEffortMedium object`941942 Medium effort. Balances latency and reasoning depth.943944 - `type: "medium"`945946 - `BetaManagedAgentsEffortHigh object`947948 High effort. Favors reasoning depth.949950 - `type: "high"`951952 - `BetaManagedAgentsEffortXhigh object`953954 Extra-high effort. Not all models accept this level.955956 - `type: "xhigh"`957958 - `BetaManagedAgentsEffortMax object`959960 Maximum effort. Favors reasoning depth over latency.961962 - `type: "max"`963964 - `inference_geo: optional string`965966 Geographic region for model inference. When unset, requests fall through to the workspace's default_inference_geo.967968 - `speed: optional "standard" or "fast"`969970 Inference speed mode. `fast` provides significantly faster output token generation at premium pricing. Not all models support `fast`; invalid combinations are rejected at create time.971972 - `"standard"`973974 - `"fast"`975976 - `multiagent: BetaManagedAgentsMultiagent or null`977978 Resolved coordinator topology with a concrete agent roster.979980 - `type: "coordinator"`981982 - `agents: array of BetaManagedAgentsAgentReference or BetaManagedAgentsAdvisor`983984 Agents the coordinator may spawn as session threads, each resolved to a specific version.985986 - `BetaManagedAgentsAgentReference object`987988 A resolved agent reference with a concrete version.989990 - `type: "agent"`991992 - `id: string`993994 - `version: number`995996 format: int32997998 - `BetaManagedAgentsAdvisor object`9991000 Platform advisor roster entry: a model the session's primary thread may consult mid-turn.10011002 - `type: "advisor"`10031004 - `model: string`10051006 The advisor model id.10071008 - `name: string`10091010 - `skills: array of BetaManagedAgentsAnthropicSkill or BetaManagedAgentsCustomSkill`10111012 - `BetaManagedAgentsAnthropicSkill object`10131014 A resolved Anthropic-managed skill.10151016 - `type: "anthropic"`10171018 - `skill_id: string`10191020 - `version: string`10211022 - `BetaManagedAgentsCustomSkill object`10231024 A resolved user-created custom skill.10251026 - `type: "custom"`10271028 - `skill_id: string`10291030 - `version: string`10311032 - `system: string or null`10331034 - `tools: array of BetaManagedAgentsAgentToolset20260401 or BetaManagedAgentsMCPToolset or BetaManagedAgentsCustomTool`10351036 - `BetaManagedAgentsAgentToolset20260401 object`10371038 - `type: "agent_toolset_20260401"`10391040 - `configs: array of BetaManagedAgentsAgentToolConfig`10411042 - `BetaManagedAgentsBashToolConfig object`10431044 Configuration for the bash tool.10451046 - `type: "bash"`10471048 - `enabled: boolean`10491050 - `name: "bash"`10511052 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`10531054 Permission policy for tool execution.10551056 - `BetaManagedAgentsAlwaysAllowPolicy object`10571058 Tool calls are automatically approved without user confirmation.10591060 - `type: "always_allow"`10611062 - `BetaManagedAgentsAlwaysAskPolicy object`10631064 Tool calls require user confirmation before execution.10651066 - `type: "always_ask"`10671068 - `BetaManagedAgentsAutoPolicy object`10691070 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.10711072 - `type: "auto"`10731074 - `BetaManagedAgentsEditToolConfig object`10751076 Configuration for the edit tool.10771078 - `type: "edit"`10791080 - `enabled: boolean`10811082 - `name: "edit"`10831084 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`10851086 Permission policy for tool execution.10871088 - `BetaManagedAgentsAlwaysAllowPolicy object`10891090 Tool calls are automatically approved without user confirmation.10911092 - `BetaManagedAgentsAlwaysAskPolicy object`10931094 Tool calls require user confirmation before execution.10951096 - `BetaManagedAgentsAutoPolicy object`10971098 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.10991100 - `BetaManagedAgentsReadToolConfig object`11011102 Configuration for the read tool.11031104 - `type: "read"`11051106 - `enabled: boolean`11071108 - `name: "read"`11091110 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`11111112 Permission policy for tool execution.11131114 - `BetaManagedAgentsAlwaysAllowPolicy object`11151116 Tool calls are automatically approved without user confirmation.11171118 - `BetaManagedAgentsAlwaysAskPolicy object`11191120 Tool calls require user confirmation before execution.11211122 - `BetaManagedAgentsAutoPolicy object`11231124 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.11251126 - `BetaManagedAgentsWriteToolConfig object`11271128 Configuration for the write tool.11291130 - `type: "write"`11311132 - `enabled: boolean`11331134 - `name: "write"`11351136 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`11371138 Permission policy for tool execution.11391140 - `BetaManagedAgentsAlwaysAllowPolicy object`11411142 Tool calls are automatically approved without user confirmation.11431144 - `BetaManagedAgentsAlwaysAskPolicy object`11451146 Tool calls require user confirmation before execution.11471148 - `BetaManagedAgentsAutoPolicy object`11491150 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.11511152 - `BetaManagedAgentsGlobToolConfig object`11531154 Configuration for the glob tool.11551156 - `type: "glob"`11571158 - `enabled: boolean`11591160 - `name: "glob"`11611162 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`11631164 Permission policy for tool execution.11651166 - `BetaManagedAgentsAlwaysAllowPolicy object`11671168 Tool calls are automatically approved without user confirmation.11691170 - `BetaManagedAgentsAlwaysAskPolicy object`11711172 Tool calls require user confirmation before execution.11731174 - `BetaManagedAgentsAutoPolicy object`11751176 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.11771178 - `BetaManagedAgentsGrepToolConfig object`11791180 Configuration for the grep tool.11811182 - `type: "grep"`11831184 - `enabled: boolean`11851186 - `name: "grep"`11871188 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`11891190 Permission policy for tool execution.11911192 - `BetaManagedAgentsAlwaysAllowPolicy object`11931194 Tool calls are automatically approved without user confirmation.11951196 - `BetaManagedAgentsAlwaysAskPolicy object`11971198 Tool calls require user confirmation before execution.11991200 - `BetaManagedAgentsAutoPolicy object`12011202 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.12031204 - `BetaManagedAgentsWebFetchToolConfig object`12051206 Configuration for the web_fetch tool.12071208 - `type: "web_fetch"`12091210 - `enabled: boolean`12111212 - `name: "web_fetch"`12131214 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`12151216 Permission policy for tool execution.12171218 - `BetaManagedAgentsAlwaysAllowPolicy object`12191220 Tool calls are automatically approved without user confirmation.12211222 - `BetaManagedAgentsAlwaysAskPolicy object`12231224 Tool calls require user confirmation before execution.12251226 - `BetaManagedAgentsAutoPolicy object`12271228 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.12291230 - `allowed_domains: optional array of string`12311232 - `blocked_domains: optional array of string`12331234 - `max_content_tokens: optional number or null`12351236 format: int3212371238 - `BetaManagedAgentsWebSearchToolConfig object`12391240 Configuration for the web_search tool.12411242 - `type: "web_search"`12431244 - `enabled: boolean`12451246 - `name: "web_search"`12471248 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`12491250 Permission policy for tool execution.12511252 - `BetaManagedAgentsAlwaysAllowPolicy object`12531254 Tool calls are automatically approved without user confirmation.12551256 - `BetaManagedAgentsAlwaysAskPolicy object`12571258 Tool calls require user confirmation before execution.12591260 - `BetaManagedAgentsAutoPolicy object`12611262 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.12631264 - `allowed_domains: optional array of string`12651266 - `blocked_domains: optional array of string`12671268 - `user_location: optional BetaManagedAgentsUserLocation or null`12691270 Approximate user location for search result localization.12711272 - `type: "approximate"`12731274 Location precision. Only "approximate" is supported.12751276 - `city: optional string or null`12771278 City name.12791280 minLength: 1, maxLength: 25512811282 - `country: optional string or null`12831284 Two-letter ISO 3166-1 country code, uppercase.12851286 - `region: optional string or null`12871288 Region or state name.12891290 minLength: 1, maxLength: 25512911292 - `timezone: optional string or null`12931294 IANA timezone identifier, e.g. "America/Los_Angeles".12951296 minLength: 1, maxLength: 25512971298 - `default_config: BetaManagedAgentsAgentToolsetDefaultConfig`12991300 Resolved default configuration for agent tools.13011302 - `enabled: boolean`13031304 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`13051306 Permission policy for tool execution.13071308 - `BetaManagedAgentsAlwaysAllowPolicy object`13091310 Tool calls are automatically approved without user confirmation.13111312 - `BetaManagedAgentsAlwaysAskPolicy object`13131314 Tool calls require user confirmation before execution.13151316 - `BetaManagedAgentsAutoPolicy object`13171318 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.13191320 - `BetaManagedAgentsMCPToolset object`13211322 - `type: "mcp_toolset"`13231324 - `configs: array of BetaManagedAgentsMCPToolConfig`13251326 - `enabled: boolean`13271328 - `name: string`13291330 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`13311332 Permission policy for tool execution.13331334 - `BetaManagedAgentsAlwaysAllowPolicy object`13351336 Tool calls are automatically approved without user confirmation.13371338 - `BetaManagedAgentsAlwaysAskPolicy object`13391340 Tool calls require user confirmation before execution.13411342 - `BetaManagedAgentsAutoPolicy object`13431344 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.13451346 - `default_config: BetaManagedAgentsMCPToolsetDefaultConfig`13471348 Resolved default configuration for all tools from an MCP server.13491350 - `enabled: boolean`13511352 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`13531354 Permission policy for tool execution.13551356 - `BetaManagedAgentsAlwaysAllowPolicy object`13571358 Tool calls are automatically approved without user confirmation.13591360 - `BetaManagedAgentsAlwaysAskPolicy object`13611362 Tool calls require user confirmation before execution.13631364 - `BetaManagedAgentsAutoPolicy object`13651366 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.13671368 - `mcp_server_name: string`13691370 - `BetaManagedAgentsCustomTool object`13711372 A custom tool as returned in API responses.13731374 - `type: "custom"`13751376 - `description: string`13771378 - `input_schema: BetaManagedAgentsCustomToolInputSchema`13791380 JSON Schema for custom tool input parameters.13811382 - `type: "object"`13831384 - `properties: optional map[unknown] or null`13851386 - `required: optional array of string or null`13871388 - `name: string`13891390 - `updated_at: string`13911392 A timestamp in RFC 3339 format13931394 format: date-time13951396 - `version: number`13971398 The agent's current version. Starts at 1 and increments when the agent is modified.13991400 format: int3214011402### Example14031404```bash1405curl https://api.anthropic.com/v1/agents \1406 -H 'Content-Type: application/json' \1407 -H 'anthropic-version: 2023-06-01' \1408 -H 'anthropic-beta: managed-agents-2026-04-01' \1409 -H "X-Api-Key: $ANTHROPIC_API_KEY" \1410 -d '{1411 "model": "claude-opus-5",1412 "name": "My First Agent",1413 "description": "A general-purpose starter agent.",1414 "metadata": {1415 "foo": "bar"1416 },1417 "system": "You are a general-purpose agent that can research, write code, run commands, and use connected tools to complete the user'\''s task end to end.",1418 "tools": [1419 {1420 "type": "agent_toolset_20260401"1421 }1422 ]1423 }'1424```14251426#### Response (200)14271428```json1429{1430 "id": "agent_011CZkYpogX7uDKUyvBTophP",1431 "archived_at": null,1432 "created_at": "2026-03-15T10:00:00Z",1433 "description": "A general-purpose starter agent.",1434 "mcp_servers": [1435 {1436 "name": "example-mcp",1437 "type": "url",1438 "url": "https://example-server.modelcontextprotocol.io/sse"1439 }1440 ],1441 "metadata": {1442 "foo": "bar"1443 },1444 "model": {1445 "id": "claude-opus-5",1446 "effort": {1447 "type": "low"1448 },1449 "inference_geo": "inference_geo",1450 "speed": "standard"1451 },1452 "multiagent": {1453 "agents": [1454 {1455 "id": "agent_011CZkYqphY8vELVzwCUpqiQ",1456 "type": "agent",1457 "version": 11458 }1459 ],1460 "type": "coordinator"1461 },1462 "name": "My First Agent",1463 "skills": [1464 {1465 "skill_id": "xlsx",1466 "type": "anthropic",1467 "version": "1"1468 },1469 {1470 "skill_id": "skill_011CZkZFNu9hAbo3jZPRgTlx",1471 "type": "custom",1472 "version": "2"1473 }1474 ],1475 "system": "You are a general-purpose agent that can research, write code, run commands, and use connected tools to complete the user's task end to end.",1476 "tools": [1477 {1478 "configs": [1479 {1480 "enabled": true,1481 "name": "bash",1482 "permission_policy": {1483 "type": "always_allow"1484 },1485 "type": "bash"1486 }1487 ],1488 "default_config": {1489 "enabled": true,1490 "permission_policy": {1491 "type": "always_ask"1492 }1493 },1494 "type": "agent_toolset_20260401"1495 }1496 ],1497 "type": "agent",1498 "updated_at": "2026-03-15T10:00:00Z",1499 "version": 11500}1501```15021503## List Agents15041505**GET** `/v1/agents`15061507List Agents15081509### Query parameters15101511- `"created_at[gte]": optional string`15121513 Return agents created at or after this time (inclusive).15141515 format: date-time15161517- `"created_at[lte]": optional string`15181519 Return agents created at or before this time (inclusive).15201521 format: date-time15221523- `include_archived: optional boolean`15241525 Include archived agents in results. Defaults to false.15261527- `limit: optional number`15281529 Maximum results per page. Default 20, maximum 100.15301531 format: int3215321533- `page: optional string`15341535 Opaque pagination cursor from a previous response.15361537### Headers15381539- `"anthropic-beta": optional array of AnthropicBeta`15401541 Optional header to specify the beta version(s) you want to use.15421543 - `string`15441545 - `"message-batches-2024-09-24" or "prompt-caching-2024-07-31" or "computer-use-2024-10-22" or 43 more`15461547 - `"message-batches-2024-09-24"`15481549 - `"prompt-caching-2024-07-31"`15501551 - `"computer-use-2024-10-22"`15521553 - `"computer-use-2025-01-24"`15541555 - `"pdfs-2024-09-25"`15561557 - `"token-counting-2024-11-01"`15581559 - `"token-efficient-tools-2025-02-19"`15601561 - `"output-128k-2025-02-19"`15621563 - `"files-api-2025-04-14"`15641565 - `"mcp-client-2025-04-04"`15661567 - `"mcp-client-2025-11-20"`15681569 - `"dev-full-thinking-2025-05-14"`15701571 - `"interleaved-thinking-2025-05-14"`15721573 - `"code-execution-2025-05-22"`15741575 - `"extended-cache-ttl-2025-04-11"`15761577 - `"context-1m-2025-08-07"`15781579 - `"context-management-2025-06-27"`15801581 - `"model-context-window-exceeded-2025-08-26"`15821583 - `"skills-2025-10-02"`15841585 - `"fast-mode-2026-02-01"`15861587 - `"output-300k-2026-03-24"`15881589 - `"user-profiles-2026-03-24"`15901591 - `"user-profiles-2026-08-18"`15921593 - `"user-profiles-2026-09-04"`15941595 - `"advisor-tool-2026-03-01"`15961597 - `"managed-agents-2026-04-01"`15981599 - `"cache-diagnosis-2026-04-07"`16001601 - `"dreaming-2026-04-21"`16021603 - `"thinking-token-count-2026-05-13"`16041605 - `"server-side-fallback-2026-06-01"`16061607 - `"server-side-fallback-2026-07-01"`16081609 - `"fallback-credit-2026-06-01"`16101611 - `"fallback-credit-2026-07-01"`16121613 - `"agent-memory-2026-07-22"`16141615 - `"mid-conversation-tool-changes-2026-07-01"`16161617 - `"compact-2026-01-12"`16181619 - `"computer-use-2025-11-24"`16201621 - `"mcp-tunnels-2026-06-22"`16221623 - `"structured-outputs-2025-11-13"`16241625 - `"task-budgets-2026-03-13"`16261627 - `"thinking-display-updates-2026-08-18"`16281629 - `"ce-user-management-2026-07-13"`16301631 - `"mid-conversation-output-config-2026-07-01"`16321633 - `"thinking-binding-controls-2026-08-01"`16341635 - `"mid-conversation-system-clear-at-2026-08-21"`16361637 - `"compact-2026-09-04"`16381639- `"anthropic-workspace-id": optional string`16401641### Returns16421643- `data: array of BetaManagedAgentsAgent`16441645 List of agents.16461647 - `type: "agent"`16481649 - `id: string`16501651 - `archived_at: string or null`16521653 A timestamp in RFC 3339 format16541655 format: date-time16561657 - `created_at: string`16581659 A timestamp in RFC 3339 format16601661 format: date-time16621663 - `description: string or null`16641665 - `mcp_servers: array of BetaManagedAgentsMCPServerURLDefinition`16661667 - `type: "url"`16681669 - `name: string`16701671 - `url: string`16721673 - `metadata: map[string]`16741675 - `model: BetaManagedAgentsModelConfig`16761677 Model identifier and configuration.16781679 - `id: BetaManagedAgentsModel`16801681 The model that will power your agent.16821683 See [models](https://docs.anthropic.com/en/docs/models-overview) for additional details and options.16841685 - `"claude-fable-5-1" or "claude-sonnet-5" or "claude-fable-5" or 11 more`16861687 The model that will power your agent.16881689 See [models](https://docs.anthropic.com/en/docs/models-overview) for additional details and options.16901691 - `"claude-fable-5-1"`16921693 Frontier intelligence for ambitious tasks across coding, scientific discovery, and enterprise workflows16941695 - `"claude-sonnet-5"`16961697 High-performance model for coding and agents16981699 - `"claude-fable-5"`17001701 Next generation of intelligence for the hardest knowledge work and coding problems17021703 - `"claude-opus-5"`17041705 Powerful intelligence for long-running agents and coding17061707 - `"claude-opus-4-8"`17081709 Powerful intelligence for long-running agents and coding17101711 - `"claude-opus-4-7"`17121713 Powerful intelligence for long-running agents and coding17141715 - `"claude-opus-4-6"`17161717 Powerful intelligence for long-running agents and coding17181719 - `"claude-sonnet-4-6"`17201721 Best combination of speed and intelligence17221723 - `"claude-haiku-4-5"`17241725 Fastest model with near-frontier intelligence17261727 - `"claude-haiku-4-5-20251001"`17281729 Fastest model with near-frontier intelligence17301731 - `"claude-opus-4-5"`17321733 Powerful intelligence for long-running agents and coding17341735 - `"claude-opus-4-5-20251101"`17361737 Powerful intelligence for long-running agents and coding17381739 - `"claude-sonnet-4-5"`17401741 High-performance model for agents and coding17421743 - `"claude-sonnet-4-5-20250929"`17441745 High-performance model for agents and coding17461747 - `string`17481749 - `effort: optional BetaManagedAgentsEffortLow or BetaManagedAgentsEffortMedium or BetaManagedAgentsEffortHigh or 2 more`17501751 How hard Claude works on each turn. Sets `output_config.effort` on every Messages call the session makes.17521753 - `BetaManagedAgentsEffortLow object`17541755 Low effort. Favors latency over reasoning depth.17561757 - `type: "low"`17581759 - `BetaManagedAgentsEffortMedium object`17601761 Medium effort. Balances latency and reasoning depth.17621763 - `type: "medium"`17641765 - `BetaManagedAgentsEffortHigh object`17661767 High effort. Favors reasoning depth.17681769 - `type: "high"`17701771 - `BetaManagedAgentsEffortXhigh object`17721773 Extra-high effort. Not all models accept this level.17741775 - `type: "xhigh"`17761777 - `BetaManagedAgentsEffortMax object`17781779 Maximum effort. Favors reasoning depth over latency.17801781 - `type: "max"`17821783 - `inference_geo: optional string`17841785 Geographic region for model inference. When unset, requests fall through to the workspace's default_inference_geo.17861787 - `speed: optional "standard" or "fast"`17881789 Inference speed mode. `fast` provides significantly faster output token generation at premium pricing. Not all models support `fast`; invalid combinations are rejected at create time.17901791 - `"standard"`17921793 - `"fast"`17941795 - `multiagent: BetaManagedAgentsMultiagent or null`17961797 Resolved coordinator topology with a concrete agent roster.17981799 - `type: "coordinator"`18001801 - `agents: array of BetaManagedAgentsAgentReference or BetaManagedAgentsAdvisor`18021803 Agents the coordinator may spawn as session threads, each resolved to a specific version.18041805 - `BetaManagedAgentsAgentReference object`18061807 A resolved agent reference with a concrete version.18081809 - `type: "agent"`18101811 - `id: string`18121813 - `version: number`18141815 format: int3218161817 - `BetaManagedAgentsAdvisor object`18181819 Platform advisor roster entry: a model the session's primary thread may consult mid-turn.18201821 - `type: "advisor"`18221823 - `model: string`18241825 The advisor model id.18261827 - `name: string`18281829 - `skills: array of BetaManagedAgentsAnthropicSkill or BetaManagedAgentsCustomSkill`18301831 - `BetaManagedAgentsAnthropicSkill object`18321833 A resolved Anthropic-managed skill.18341835 - `type: "anthropic"`18361837 - `skill_id: string`18381839 - `version: string`18401841 - `BetaManagedAgentsCustomSkill object`18421843 A resolved user-created custom skill.18441845 - `type: "custom"`18461847 - `skill_id: string`18481849 - `version: string`18501851 - `system: string or null`18521853 - `tools: array of BetaManagedAgentsAgentToolset20260401 or BetaManagedAgentsMCPToolset or BetaManagedAgentsCustomTool`18541855 - `BetaManagedAgentsAgentToolset20260401 object`18561857 - `type: "agent_toolset_20260401"`18581859 - `configs: array of BetaManagedAgentsAgentToolConfig`18601861 - `BetaManagedAgentsBashToolConfig object`18621863 Configuration for the bash tool.18641865 - `type: "bash"`18661867 - `enabled: boolean`18681869 - `name: "bash"`18701871 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`18721873 Permission policy for tool execution.18741875 - `BetaManagedAgentsAlwaysAllowPolicy object`18761877 Tool calls are automatically approved without user confirmation.18781879 - `type: "always_allow"`18801881 - `BetaManagedAgentsAlwaysAskPolicy object`18821883 Tool calls require user confirmation before execution.18841885 - `type: "always_ask"`18861887 - `BetaManagedAgentsAutoPolicy object`18881889 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.18901891 - `type: "auto"`18921893 - `BetaManagedAgentsEditToolConfig object`18941895 Configuration for the edit tool.18961897 - `type: "edit"`18981899 - `enabled: boolean`19001901 - `name: "edit"`19021903 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`19041905 Permission policy for tool execution.19061907 - `BetaManagedAgentsAlwaysAllowPolicy object`19081909 Tool calls are automatically approved without user confirmation.19101911 - `BetaManagedAgentsAlwaysAskPolicy object`19121913 Tool calls require user confirmation before execution.19141915 - `BetaManagedAgentsAutoPolicy object`19161917 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.19181919 - `BetaManagedAgentsReadToolConfig object`19201921 Configuration for the read tool.19221923 - `type: "read"`19241925 - `enabled: boolean`19261927 - `name: "read"`19281929 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`19301931 Permission policy for tool execution.19321933 - `BetaManagedAgentsAlwaysAllowPolicy object`19341935 Tool calls are automatically approved without user confirmation.19361937 - `BetaManagedAgentsAlwaysAskPolicy object`19381939 Tool calls require user confirmation before execution.19401941 - `BetaManagedAgentsAutoPolicy object`19421943 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.19441945 - `BetaManagedAgentsWriteToolConfig object`19461947 Configuration for the write tool.19481949 - `type: "write"`19501951 - `enabled: boolean`19521953 - `name: "write"`19541955 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`19561957 Permission policy for tool execution.19581959 - `BetaManagedAgentsAlwaysAllowPolicy object`19601961 Tool calls are automatically approved without user confirmation.19621963 - `BetaManagedAgentsAlwaysAskPolicy object`19641965 Tool calls require user confirmation before execution.19661967 - `BetaManagedAgentsAutoPolicy object`19681969 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.19701971 - `BetaManagedAgentsGlobToolConfig object`19721973 Configuration for the glob tool.19741975 - `type: "glob"`19761977 - `enabled: boolean`19781979 - `name: "glob"`19801981 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`19821983 Permission policy for tool execution.19841985 - `BetaManagedAgentsAlwaysAllowPolicy object`19861987 Tool calls are automatically approved without user confirmation.19881989 - `BetaManagedAgentsAlwaysAskPolicy object`19901991 Tool calls require user confirmation before execution.19921993 - `BetaManagedAgentsAutoPolicy object`19941995 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.19961997 - `BetaManagedAgentsGrepToolConfig object`19981999 Configuration for the grep tool.20002001 - `type: "grep"`20022003 - `enabled: boolean`20042005 - `name: "grep"`20062007 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`20082009 Permission policy for tool execution.20102011 - `BetaManagedAgentsAlwaysAllowPolicy object`20122013 Tool calls are automatically approved without user confirmation.20142015 - `BetaManagedAgentsAlwaysAskPolicy object`20162017 Tool calls require user confirmation before execution.20182019 - `BetaManagedAgentsAutoPolicy object`20202021 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.20222023 - `BetaManagedAgentsWebFetchToolConfig object`20242025 Configuration for the web_fetch tool.20262027 - `type: "web_fetch"`20282029 - `enabled: boolean`20302031 - `name: "web_fetch"`20322033 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`20342035 Permission policy for tool execution.20362037 - `BetaManagedAgentsAlwaysAllowPolicy object`20382039 Tool calls are automatically approved without user confirmation.20402041 - `BetaManagedAgentsAlwaysAskPolicy object`20422043 Tool calls require user confirmation before execution.20442045 - `BetaManagedAgentsAutoPolicy object`20462047 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.20482049 - `allowed_domains: optional array of string`20502051 - `blocked_domains: optional array of string`20522053 - `max_content_tokens: optional number or null`20542055 format: int3220562057 - `BetaManagedAgentsWebSearchToolConfig object`20582059 Configuration for the web_search tool.20602061 - `type: "web_search"`20622063 - `enabled: boolean`20642065 - `name: "web_search"`20662067 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`20682069 Permission policy for tool execution.20702071 - `BetaManagedAgentsAlwaysAllowPolicy object`20722073 Tool calls are automatically approved without user confirmation.20742075 - `BetaManagedAgentsAlwaysAskPolicy object`20762077 Tool calls require user confirmation before execution.20782079 - `BetaManagedAgentsAutoPolicy object`20802081 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.20822083 - `allowed_domains: optional array of string`20842085 - `blocked_domains: optional array of string`20862087 - `user_location: optional BetaManagedAgentsUserLocation or null`20882089 Approximate user location for search result localization.20902091 - `type: "approximate"`20922093 Location precision. Only "approximate" is supported.20942095 - `city: optional string or null`20962097 City name.20982099 minLength: 1, maxLength: 25521002101 - `country: optional string or null`21022103 Two-letter ISO 3166-1 country code, uppercase.21042105 - `region: optional string or null`21062107 Region or state name.21082109 minLength: 1, maxLength: 25521102111 - `timezone: optional string or null`21122113 IANA timezone identifier, e.g. "America/Los_Angeles".21142115 minLength: 1, maxLength: 25521162117 - `default_config: BetaManagedAgentsAgentToolsetDefaultConfig`21182119 Resolved default configuration for agent tools.21202121 - `enabled: boolean`21222123 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`21242125 Permission policy for tool execution.21262127 - `BetaManagedAgentsAlwaysAllowPolicy object`21282129 Tool calls are automatically approved without user confirmation.21302131 - `BetaManagedAgentsAlwaysAskPolicy object`21322133 Tool calls require user confirmation before execution.21342135 - `BetaManagedAgentsAutoPolicy object`21362137 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.21382139 - `BetaManagedAgentsMCPToolset object`21402141 - `type: "mcp_toolset"`21422143 - `configs: array of BetaManagedAgentsMCPToolConfig`21442145 - `enabled: boolean`21462147 - `name: string`21482149 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`21502151 Permission policy for tool execution.21522153 - `BetaManagedAgentsAlwaysAllowPolicy object`21542155 Tool calls are automatically approved without user confirmation.21562157 - `BetaManagedAgentsAlwaysAskPolicy object`21582159 Tool calls require user confirmation before execution.21602161 - `BetaManagedAgentsAutoPolicy object`21622163 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.21642165 - `default_config: BetaManagedAgentsMCPToolsetDefaultConfig`21662167 Resolved default configuration for all tools from an MCP server.21682169 - `enabled: boolean`21702171 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`21722173 Permission policy for tool execution.21742175 - `BetaManagedAgentsAlwaysAllowPolicy object`21762177 Tool calls are automatically approved without user confirmation.21782179 - `BetaManagedAgentsAlwaysAskPolicy object`21802181 Tool calls require user confirmation before execution.21822183 - `BetaManagedAgentsAutoPolicy object`21842185 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.21862187 - `mcp_server_name: string`21882189 - `BetaManagedAgentsCustomTool object`21902191 A custom tool as returned in API responses.21922193 - `type: "custom"`21942195 - `description: string`21962197 - `input_schema: BetaManagedAgentsCustomToolInputSchema`21982199 JSON Schema for custom tool input parameters.22002201 - `type: "object"`22022203 - `properties: optional map[unknown] or null`22042205 - `required: optional array of string or null`22062207 - `name: string`22082209 - `updated_at: string`22102211 A timestamp in RFC 3339 format22122213 format: date-time22142215 - `version: number`22162217 The agent's current version. Starts at 1 and increments when the agent is modified.22182219 format: int3222202221- `next_page: optional string or null`22222223 Opaque cursor for the next page. Null when no more results.22242225### Example22262227```bash2228curl https://api.anthropic.com/v1/agents \2229 -H 'anthropic-version: 2023-06-01' \2230 -H 'anthropic-beta: managed-agents-2026-04-01' \2231 -H "X-Api-Key: $ANTHROPIC_API_KEY"2232```22332234#### Response (200)22352236```json2237{2238 "data": [2239 {2240 "id": "agent_011CZkYpogX7uDKUyvBTophP",2241 "archived_at": null,2242 "created_at": "2026-03-15T10:00:00Z",2243 "description": "A general-purpose starter agent.",2244 "mcp_servers": [2245 {2246 "name": "example-mcp",2247 "type": "url",2248 "url": "https://example-server.modelcontextprotocol.io/sse"2249 }2250 ],2251 "metadata": {2252 "foo": "bar"2253 },2254 "model": {2255 "id": "claude-opus-5",2256 "effort": {2257 "type": "low"2258 },2259 "inference_geo": "inference_geo",2260 "speed": "standard"2261 },2262 "multiagent": {2263 "agents": [2264 {2265 "id": "agent_011CZkYqphY8vELVzwCUpqiQ",2266 "type": "agent",2267 "version": 12268 }2269 ],2270 "type": "coordinator"2271 },2272 "name": "My First Agent",2273 "skills": [2274 {2275 "skill_id": "xlsx",2276 "type": "anthropic",2277 "version": "1"2278 },2279 {2280 "skill_id": "skill_011CZkZFNu9hAbo3jZPRgTlx",2281 "type": "custom",2282 "version": "2"2283 }2284 ],2285 "system": "You are a general-purpose agent that can research, write code, run commands, and use connected tools to complete the user's task end to end.",2286 "tools": [2287 {2288 "configs": [2289 {2290 "enabled": true,2291 "name": "bash",2292 "permission_policy": {2293 "type": "always_allow"2294 },2295 "type": "bash"2296 }2297 ],2298 "default_config": {2299 "enabled": true,2300 "permission_policy": {2301 "type": "always_ask"2302 }2303 },2304 "type": "agent_toolset_20260401"2305 }2306 ],2307 "type": "agent",2308 "updated_at": "2026-03-15T10:00:00Z",2309 "version": 12310 }2311 ],2312 "next_page": "next_page"2313}2314```23152316## Get Agent23172318**GET** `/v1/agents/{agent_id}`23192320Get Agent23212322### Path parameters23232324- `agent_id: string`23252326### Query parameters23272328- `version: optional number`23292330 Agent version. Omit for the most recent version. Must be at least 1 if specified.23312332 format: int3223332334### Headers23352336- `"anthropic-beta": optional array of AnthropicBeta`23372338 Optional header to specify the beta version(s) you want to use.23392340 - `string`23412342 - `"message-batches-2024-09-24" or "prompt-caching-2024-07-31" or "computer-use-2024-10-22" or 43 more`23432344 - `"message-batches-2024-09-24"`23452346 - `"prompt-caching-2024-07-31"`23472348 - `"computer-use-2024-10-22"`23492350 - `"computer-use-2025-01-24"`23512352 - `"pdfs-2024-09-25"`23532354 - `"token-counting-2024-11-01"`23552356 - `"token-efficient-tools-2025-02-19"`23572358 - `"output-128k-2025-02-19"`23592360 - `"files-api-2025-04-14"`23612362 - `"mcp-client-2025-04-04"`23632364 - `"mcp-client-2025-11-20"`23652366 - `"dev-full-thinking-2025-05-14"`23672368 - `"interleaved-thinking-2025-05-14"`23692370 - `"code-execution-2025-05-22"`23712372 - `"extended-cache-ttl-2025-04-11"`23732374 - `"context-1m-2025-08-07"`23752376 - `"context-management-2025-06-27"`23772378 - `"model-context-window-exceeded-2025-08-26"`23792380 - `"skills-2025-10-02"`23812382 - `"fast-mode-2026-02-01"`23832384 - `"output-300k-2026-03-24"`23852386 - `"user-profiles-2026-03-24"`23872388 - `"user-profiles-2026-08-18"`23892390 - `"user-profiles-2026-09-04"`23912392 - `"advisor-tool-2026-03-01"`23932394 - `"managed-agents-2026-04-01"`23952396 - `"cache-diagnosis-2026-04-07"`23972398 - `"dreaming-2026-04-21"`23992400 - `"thinking-token-count-2026-05-13"`24012402 - `"server-side-fallback-2026-06-01"`24032404 - `"server-side-fallback-2026-07-01"`24052406 - `"fallback-credit-2026-06-01"`24072408 - `"fallback-credit-2026-07-01"`24092410 - `"agent-memory-2026-07-22"`24112412 - `"mid-conversation-tool-changes-2026-07-01"`24132414 - `"compact-2026-01-12"`24152416 - `"computer-use-2025-11-24"`24172418 - `"mcp-tunnels-2026-06-22"`24192420 - `"structured-outputs-2025-11-13"`24212422 - `"task-budgets-2026-03-13"`24232424 - `"thinking-display-updates-2026-08-18"`24252426 - `"ce-user-management-2026-07-13"`24272428 - `"mid-conversation-output-config-2026-07-01"`24292430 - `"thinking-binding-controls-2026-08-01"`24312432 - `"mid-conversation-system-clear-at-2026-08-21"`24332434 - `"compact-2026-09-04"`24352436- `"anthropic-workspace-id": optional string`24372438### Returns24392440- `BetaManagedAgentsAgent object`24412442 A Managed Agents `agent`.24432444 - `type: "agent"`24452446 - `id: string`24472448 - `archived_at: string or null`24492450 A timestamp in RFC 3339 format24512452 format: date-time24532454 - `created_at: string`24552456 A timestamp in RFC 3339 format24572458 format: date-time24592460 - `description: string or null`24612462 - `mcp_servers: array of BetaManagedAgentsMCPServerURLDefinition`24632464 - `type: "url"`24652466 - `name: string`24672468 - `url: string`24692470 - `metadata: map[string]`24712472 - `model: BetaManagedAgentsModelConfig`24732474 Model identifier and configuration.24752476 - `id: BetaManagedAgentsModel`24772478 The model that will power your agent.24792480 See [models](https://docs.anthropic.com/en/docs/models-overview) for additional details and options.24812482 - `"claude-fable-5-1" or "claude-sonnet-5" or "claude-fable-5" or 11 more`24832484 The model that will power your agent.24852486 See [models](https://docs.anthropic.com/en/docs/models-overview) for additional details and options.24872488 - `"claude-fable-5-1"`24892490 Frontier intelligence for ambitious tasks across coding, scientific discovery, and enterprise workflows24912492 - `"claude-sonnet-5"`24932494 High-performance model for coding and agents24952496 - `"claude-fable-5"`24972498 Next generation of intelligence for the hardest knowledge work and coding problems24992500 - `"claude-opus-5"`25012502 Powerful intelligence for long-running agents and coding25032504 - `"claude-opus-4-8"`25052506 Powerful intelligence for long-running agents and coding25072508 - `"claude-opus-4-7"`25092510 Powerful intelligence for long-running agents and coding25112512 - `"claude-opus-4-6"`25132514 Powerful intelligence for long-running agents and coding25152516 - `"claude-sonnet-4-6"`25172518 Best combination of speed and intelligence25192520 - `"claude-haiku-4-5"`25212522 Fastest model with near-frontier intelligence25232524 - `"claude-haiku-4-5-20251001"`25252526 Fastest model with near-frontier intelligence25272528 - `"claude-opus-4-5"`25292530 Powerful intelligence for long-running agents and coding25312532 - `"claude-opus-4-5-20251101"`25332534 Powerful intelligence for long-running agents and coding25352536 - `"claude-sonnet-4-5"`25372538 High-performance model for agents and coding25392540 - `"claude-sonnet-4-5-20250929"`25412542 High-performance model for agents and coding25432544 - `string`25452546 - `effort: optional BetaManagedAgentsEffortLow or BetaManagedAgentsEffortMedium or BetaManagedAgentsEffortHigh or 2 more`25472548 How hard Claude works on each turn. Sets `output_config.effort` on every Messages call the session makes.25492550 - `BetaManagedAgentsEffortLow object`25512552 Low effort. Favors latency over reasoning depth.25532554 - `type: "low"`25552556 - `BetaManagedAgentsEffortMedium object`25572558 Medium effort. Balances latency and reasoning depth.25592560 - `type: "medium"`25612562 - `BetaManagedAgentsEffortHigh object`25632564 High effort. Favors reasoning depth.25652566 - `type: "high"`25672568 - `BetaManagedAgentsEffortXhigh object`25692570 Extra-high effort. Not all models accept this level.25712572 - `type: "xhigh"`25732574 - `BetaManagedAgentsEffortMax object`25752576 Maximum effort. Favors reasoning depth over latency.25772578 - `type: "max"`25792580 - `inference_geo: optional string`25812582 Geographic region for model inference. When unset, requests fall through to the workspace's default_inference_geo.25832584 - `speed: optional "standard" or "fast"`25852586 Inference speed mode. `fast` provides significantly faster output token generation at premium pricing. Not all models support `fast`; invalid combinations are rejected at create time.25872588 - `"standard"`25892590 - `"fast"`25912592 - `multiagent: BetaManagedAgentsMultiagent or null`25932594 Resolved coordinator topology with a concrete agent roster.25952596 - `type: "coordinator"`25972598 - `agents: array of BetaManagedAgentsAgentReference or BetaManagedAgentsAdvisor`25992600 Agents the coordinator may spawn as session threads, each resolved to a specific version.26012602 - `BetaManagedAgentsAgentReference object`26032604 A resolved agent reference with a concrete version.26052606 - `type: "agent"`26072608 - `id: string`26092610 - `version: number`26112612 format: int3226132614 - `BetaManagedAgentsAdvisor object`26152616 Platform advisor roster entry: a model the session's primary thread may consult mid-turn.26172618 - `type: "advisor"`26192620 - `model: string`26212622 The advisor model id.26232624 - `name: string`26252626 - `skills: array of BetaManagedAgentsAnthropicSkill or BetaManagedAgentsCustomSkill`26272628 - `BetaManagedAgentsAnthropicSkill object`26292630 A resolved Anthropic-managed skill.26312632 - `type: "anthropic"`26332634 - `skill_id: string`26352636 - `version: string`26372638 - `BetaManagedAgentsCustomSkill object`26392640 A resolved user-created custom skill.26412642 - `type: "custom"`26432644 - `skill_id: string`26452646 - `version: string`26472648 - `system: string or null`26492650 - `tools: array of BetaManagedAgentsAgentToolset20260401 or BetaManagedAgentsMCPToolset or BetaManagedAgentsCustomTool`26512652 - `BetaManagedAgentsAgentToolset20260401 object`26532654 - `type: "agent_toolset_20260401"`26552656 - `configs: array of BetaManagedAgentsAgentToolConfig`26572658 - `BetaManagedAgentsBashToolConfig object`26592660 Configuration for the bash tool.26612662 - `type: "bash"`26632664 - `enabled: boolean`26652666 - `name: "bash"`26672668 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`26692670 Permission policy for tool execution.26712672 - `BetaManagedAgentsAlwaysAllowPolicy object`26732674 Tool calls are automatically approved without user confirmation.26752676 - `type: "always_allow"`26772678 - `BetaManagedAgentsAlwaysAskPolicy object`26792680 Tool calls require user confirmation before execution.26812682 - `type: "always_ask"`26832684 - `BetaManagedAgentsAutoPolicy object`26852686 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.26872688 - `type: "auto"`26892690 - `BetaManagedAgentsEditToolConfig object`26912692 Configuration for the edit tool.26932694 - `type: "edit"`26952696 - `enabled: boolean`26972698 - `name: "edit"`26992700 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`27012702 Permission policy for tool execution.27032704 - `BetaManagedAgentsAlwaysAllowPolicy object`27052706 Tool calls are automatically approved without user confirmation.27072708 - `BetaManagedAgentsAlwaysAskPolicy object`27092710 Tool calls require user confirmation before execution.27112712 - `BetaManagedAgentsAutoPolicy object`27132714 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.27152716 - `BetaManagedAgentsReadToolConfig object`27172718 Configuration for the read tool.27192720 - `type: "read"`27212722 - `enabled: boolean`27232724 - `name: "read"`27252726 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`27272728 Permission policy for tool execution.27292730 - `BetaManagedAgentsAlwaysAllowPolicy object`27312732 Tool calls are automatically approved without user confirmation.27332734 - `BetaManagedAgentsAlwaysAskPolicy object`27352736 Tool calls require user confirmation before execution.27372738 - `BetaManagedAgentsAutoPolicy object`27392740 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.27412742 - `BetaManagedAgentsWriteToolConfig object`27432744 Configuration for the write tool.27452746 - `type: "write"`27472748 - `enabled: boolean`27492750 - `name: "write"`27512752 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`27532754 Permission policy for tool execution.27552756 - `BetaManagedAgentsAlwaysAllowPolicy object`27572758 Tool calls are automatically approved without user confirmation.27592760 - `BetaManagedAgentsAlwaysAskPolicy object`27612762 Tool calls require user confirmation before execution.27632764 - `BetaManagedAgentsAutoPolicy object`27652766 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.27672768 - `BetaManagedAgentsGlobToolConfig object`27692770 Configuration for the glob tool.27712772 - `type: "glob"`27732774 - `enabled: boolean`27752776 - `name: "glob"`27772778 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`27792780 Permission policy for tool execution.27812782 - `BetaManagedAgentsAlwaysAllowPolicy object`27832784 Tool calls are automatically approved without user confirmation.27852786 - `BetaManagedAgentsAlwaysAskPolicy object`27872788 Tool calls require user confirmation before execution.27892790 - `BetaManagedAgentsAutoPolicy object`27912792 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.27932794 - `BetaManagedAgentsGrepToolConfig object`27952796 Configuration for the grep tool.27972798 - `type: "grep"`27992800 - `enabled: boolean`28012802 - `name: "grep"`28032804 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`28052806 Permission policy for tool execution.28072808 - `BetaManagedAgentsAlwaysAllowPolicy object`28092810 Tool calls are automatically approved without user confirmation.28112812 - `BetaManagedAgentsAlwaysAskPolicy object`28132814 Tool calls require user confirmation before execution.28152816 - `BetaManagedAgentsAutoPolicy object`28172818 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.28192820 - `BetaManagedAgentsWebFetchToolConfig object`28212822 Configuration for the web_fetch tool.28232824 - `type: "web_fetch"`28252826 - `enabled: boolean`28272828 - `name: "web_fetch"`28292830 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`28312832 Permission policy for tool execution.28332834 - `BetaManagedAgentsAlwaysAllowPolicy object`28352836 Tool calls are automatically approved without user confirmation.28372838 - `BetaManagedAgentsAlwaysAskPolicy object`28392840 Tool calls require user confirmation before execution.28412842 - `BetaManagedAgentsAutoPolicy object`28432844 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.28452846 - `allowed_domains: optional array of string`28472848 - `blocked_domains: optional array of string`28492850 - `max_content_tokens: optional number or null`28512852 format: int3228532854 - `BetaManagedAgentsWebSearchToolConfig object`28552856 Configuration for the web_search tool.28572858 - `type: "web_search"`28592860 - `enabled: boolean`28612862 - `name: "web_search"`28632864 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`28652866 Permission policy for tool execution.28672868 - `BetaManagedAgentsAlwaysAllowPolicy object`28692870 Tool calls are automatically approved without user confirmation.28712872 - `BetaManagedAgentsAlwaysAskPolicy object`28732874 Tool calls require user confirmation before execution.28752876 - `BetaManagedAgentsAutoPolicy object`28772878 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.28792880 - `allowed_domains: optional array of string`28812882 - `blocked_domains: optional array of string`28832884 - `user_location: optional BetaManagedAgentsUserLocation or null`28852886 Approximate user location for search result localization.28872888 - `type: "approximate"`28892890 Location precision. Only "approximate" is supported.28912892 - `city: optional string or null`28932894 City name.28952896 minLength: 1, maxLength: 25528972898 - `country: optional string or null`28992900 Two-letter ISO 3166-1 country code, uppercase.29012902 - `region: optional string or null`29032904 Region or state name.29052906 minLength: 1, maxLength: 25529072908 - `timezone: optional string or null`29092910 IANA timezone identifier, e.g. "America/Los_Angeles".29112912 minLength: 1, maxLength: 25529132914 - `default_config: BetaManagedAgentsAgentToolsetDefaultConfig`29152916 Resolved default configuration for agent tools.29172918 - `enabled: boolean`29192920 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`29212922 Permission policy for tool execution.29232924 - `BetaManagedAgentsAlwaysAllowPolicy object`29252926 Tool calls are automatically approved without user confirmation.29272928 - `BetaManagedAgentsAlwaysAskPolicy object`29292930 Tool calls require user confirmation before execution.29312932 - `BetaManagedAgentsAutoPolicy object`29332934 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.29352936 - `BetaManagedAgentsMCPToolset object`29372938 - `type: "mcp_toolset"`29392940 - `configs: array of BetaManagedAgentsMCPToolConfig`29412942 - `enabled: boolean`29432944 - `name: string`29452946 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`29472948 Permission policy for tool execution.29492950 - `BetaManagedAgentsAlwaysAllowPolicy object`29512952 Tool calls are automatically approved without user confirmation.29532954 - `BetaManagedAgentsAlwaysAskPolicy object`29552956 Tool calls require user confirmation before execution.29572958 - `BetaManagedAgentsAutoPolicy object`29592960 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.29612962 - `default_config: BetaManagedAgentsMCPToolsetDefaultConfig`29632964 Resolved default configuration for all tools from an MCP server.29652966 - `enabled: boolean`29672968 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`29692970 Permission policy for tool execution.29712972 - `BetaManagedAgentsAlwaysAllowPolicy object`29732974 Tool calls are automatically approved without user confirmation.29752976 - `BetaManagedAgentsAlwaysAskPolicy object`29772978 Tool calls require user confirmation before execution.29792980 - `BetaManagedAgentsAutoPolicy object`29812982 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.29832984 - `mcp_server_name: string`29852986 - `BetaManagedAgentsCustomTool object`29872988 A custom tool as returned in API responses.29892990 - `type: "custom"`29912992 - `description: string`29932994 - `input_schema: BetaManagedAgentsCustomToolInputSchema`29952996 JSON Schema for custom tool input parameters.29972998 - `type: "object"`29993000 - `properties: optional map[unknown] or null`30013002 - `required: optional array of string or null`30033004 - `name: string`30053006 - `updated_at: string`30073008 A timestamp in RFC 3339 format30093010 format: date-time30113012 - `version: number`30133014 The agent's current version. Starts at 1 and increments when the agent is modified.30153016 format: int3230173018### Example30193020```bash3021curl https://api.anthropic.com/v1/agents/$AGENT_ID \3022 -H 'anthropic-version: 2023-06-01' \3023 -H 'anthropic-beta: managed-agents-2026-04-01' \3024 -H "X-Api-Key: $ANTHROPIC_API_KEY"3025```30263027#### Response (200)30283029```json3030{3031 "id": "agent_011CZkYpogX7uDKUyvBTophP",3032 "archived_at": null,3033 "created_at": "2026-03-15T10:00:00Z",3034 "description": "A general-purpose starter agent.",3035 "mcp_servers": [3036 {3037 "name": "example-mcp",3038 "type": "url",3039 "url": "https://example-server.modelcontextprotocol.io/sse"3040 }3041 ],3042 "metadata": {3043 "foo": "bar"3044 },3045 "model": {3046 "id": "claude-opus-5",3047 "effort": {3048 "type": "low"3049 },3050 "inference_geo": "inference_geo",3051 "speed": "standard"3052 },3053 "multiagent": {3054 "agents": [3055 {3056 "id": "agent_011CZkYqphY8vELVzwCUpqiQ",3057 "type": "agent",3058 "version": 13059 }3060 ],3061 "type": "coordinator"3062 },3063 "name": "My First Agent",3064 "skills": [3065 {3066 "skill_id": "xlsx",3067 "type": "anthropic",3068 "version": "1"3069 },3070 {3071 "skill_id": "skill_011CZkZFNu9hAbo3jZPRgTlx",3072 "type": "custom",3073 "version": "2"3074 }3075 ],3076 "system": "You are a general-purpose agent that can research, write code, run commands, and use connected tools to complete the user's task end to end.",3077 "tools": [3078 {3079 "configs": [3080 {3081 "enabled": true,3082 "name": "bash",3083 "permission_policy": {3084 "type": "always_allow"3085 },3086 "type": "bash"3087 }3088 ],3089 "default_config": {3090 "enabled": true,3091 "permission_policy": {3092 "type": "always_ask"3093 }3094 },3095 "type": "agent_toolset_20260401"3096 }3097 ],3098 "type": "agent",3099 "updated_at": "2026-03-15T10:00:00Z",3100 "version": 13101}3102```31033104## Update Agent31053106**POST** `/v1/agents/{agent_id}`31073108Update Agent31093110### Path parameters31113112- `agent_id: string`31133114### Headers31153116- `"anthropic-beta": optional array of AnthropicBeta`31173118 Optional header to specify the beta version(s) you want to use.31193120 - `string`31213122 - `"message-batches-2024-09-24" or "prompt-caching-2024-07-31" or "computer-use-2024-10-22" or 43 more`31233124 - `"message-batches-2024-09-24"`31253126 - `"prompt-caching-2024-07-31"`31273128 - `"computer-use-2024-10-22"`31293130 - `"computer-use-2025-01-24"`31313132 - `"pdfs-2024-09-25"`31333134 - `"token-counting-2024-11-01"`31353136 - `"token-efficient-tools-2025-02-19"`31373138 - `"output-128k-2025-02-19"`31393140 - `"files-api-2025-04-14"`31413142 - `"mcp-client-2025-04-04"`31433144 - `"mcp-client-2025-11-20"`31453146 - `"dev-full-thinking-2025-05-14"`31473148 - `"interleaved-thinking-2025-05-14"`31493150 - `"code-execution-2025-05-22"`31513152 - `"extended-cache-ttl-2025-04-11"`31533154 - `"context-1m-2025-08-07"`31553156 - `"context-management-2025-06-27"`31573158 - `"model-context-window-exceeded-2025-08-26"`31593160 - `"skills-2025-10-02"`31613162 - `"fast-mode-2026-02-01"`31633164 - `"output-300k-2026-03-24"`31653166 - `"user-profiles-2026-03-24"`31673168 - `"user-profiles-2026-08-18"`31693170 - `"user-profiles-2026-09-04"`31713172 - `"advisor-tool-2026-03-01"`31733174 - `"managed-agents-2026-04-01"`31753176 - `"cache-diagnosis-2026-04-07"`31773178 - `"dreaming-2026-04-21"`31793180 - `"thinking-token-count-2026-05-13"`31813182 - `"server-side-fallback-2026-06-01"`31833184 - `"server-side-fallback-2026-07-01"`31853186 - `"fallback-credit-2026-06-01"`31873188 - `"fallback-credit-2026-07-01"`31893190 - `"agent-memory-2026-07-22"`31913192 - `"mid-conversation-tool-changes-2026-07-01"`31933194 - `"compact-2026-01-12"`31953196 - `"computer-use-2025-11-24"`31973198 - `"mcp-tunnels-2026-06-22"`31993200 - `"structured-outputs-2025-11-13"`32013202 - `"task-budgets-2026-03-13"`32033204 - `"thinking-display-updates-2026-08-18"`32053206 - `"ce-user-management-2026-07-13"`32073208 - `"mid-conversation-output-config-2026-07-01"`32093210 - `"thinking-binding-controls-2026-08-01"`32113212 - `"mid-conversation-system-clear-at-2026-08-21"`32133214 - `"compact-2026-09-04"`32153216- `"anthropic-workspace-id": optional string`32173218### Body parameters32193220- `description: optional string or null`32213222 Description. Omit to preserve; send empty string or null to clear.32233224 maxLength: 204832253226- `mcp_servers: optional array of BetaManagedAgentsURLMCPServerParams or null`32273228 MCP servers. Full replacement. Omit to preserve; send empty array or `null` to clear. Names must be unique. Maximum 20. Every server must be referenced by an `mcp_toolset` in the agent's resulting `tools`; unreferenced servers are rejected. See the [MCP connector guide](https://platform.claude.com/docs/en/managed-agents/mcp-connector).32293230 - `type: "url"`32313232 - `name: string`32333234 Unique name for this server, referenced by mcp_toolset configurations. 1-255 characters.32353236 minLength: 1, maxLength: 25532373238 - `url: string`32393240 Endpoint URL for the MCP server.32413242 maxLength: 204832433244- `metadata: optional map[string] or null`32453246 Metadata patch. Set a key to a string to upsert it, or to null to delete it. Omit the field to preserve. The stored bag is limited to 16 keys (up to 64 chars each) with values up to 512 chars.32473248- `model: optional BetaManagedAgentsModel or BetaManagedAgentsModelConfigParams`32493250 Model identifier. Accepts the [model string](https://platform.claude.com/docs/en/about-claude/models/overview#latest-models-comparison), e.g. `claude-opus-5`, or a `model_config` object for additional configuration control. Omit to preserve. Cannot be cleared.32513252 - `BetaManagedAgentsModel = "claude-fable-5-1" or "claude-sonnet-5" or "claude-fable-5" or 11 more or string`32533254 The model that will power your agent.32553256 See [models](https://docs.anthropic.com/en/docs/models-overview) for additional details and options.32573258 - `"claude-fable-5-1" or "claude-sonnet-5" or "claude-fable-5" or 11 more`32593260 The model that will power your agent.32613262 See [models](https://docs.anthropic.com/en/docs/models-overview) for additional details and options.32633264 - `"claude-fable-5-1"`32653266 Frontier intelligence for ambitious tasks across coding, scientific discovery, and enterprise workflows32673268 - `"claude-sonnet-5"`32693270 High-performance model for coding and agents32713272 - `"claude-fable-5"`32733274 Next generation of intelligence for the hardest knowledge work and coding problems32753276 - `"claude-opus-5"`32773278 Powerful intelligence for long-running agents and coding32793280 - `"claude-opus-4-8"`32813282 Powerful intelligence for long-running agents and coding32833284 - `"claude-opus-4-7"`32853286 Powerful intelligence for long-running agents and coding32873288 - `"claude-opus-4-6"`32893290 Powerful intelligence for long-running agents and coding32913292 - `"claude-sonnet-4-6"`32933294 Best combination of speed and intelligence32953296 - `"claude-haiku-4-5"`32973298 Fastest model with near-frontier intelligence32993300 - `"claude-haiku-4-5-20251001"`33013302 Fastest model with near-frontier intelligence33033304 - `"claude-opus-4-5"`33053306 Powerful intelligence for long-running agents and coding33073308 - `"claude-opus-4-5-20251101"`33093310 Powerful intelligence for long-running agents and coding33113312 - `"claude-sonnet-4-5"`33133314 High-performance model for agents and coding33153316 - `"claude-sonnet-4-5-20250929"`33173318 High-performance model for agents and coding33193320 - `string`33213322 - `BetaManagedAgentsModelConfigParams object`33233324 An object that defines additional configuration control over model use33253326 - `id: BetaManagedAgentsModel`33273328 The model that will power your agent.33293330 See [models](https://docs.anthropic.com/en/docs/models-overview) for additional details and options.33313332 - `effort: optional "low" or "medium" or "high" or 2 more or BetaManagedAgentsEffortLow or BetaManagedAgentsEffortMedium or 3 more or null`33333334 How hard Claude works on each inference call. Accepts a bare level string (`"high"`) or `{"type": "high"}`. On create, omitting it resolves the per-model default; on update, omitting it leaves the stored value unchanged.33353336 - `BetaManagedAgentsEffortLevel = "low" or "medium" or "high" or 2 more`33373338 How hard Claude works on each turn. Higher levels favor reasoning depth over latency. Not all models accept every level; invalid combinations are rejected at create time.33393340 - `"low"`33413342 - `"medium"`33433344 - `"high"`33453346 - `"xhigh"`33473348 - `"max"`33493350 - `BetaManagedAgentsEffortLow object`33513352 Low effort. Favors latency over reasoning depth.33533354 - `type: "low"`33553356 - `BetaManagedAgentsEffortMedium object`33573358 Medium effort. Balances latency and reasoning depth.33593360 - `type: "medium"`33613362 - `BetaManagedAgentsEffortHigh object`33633364 High effort. Favors reasoning depth.33653366 - `type: "high"`33673368 - `BetaManagedAgentsEffortXhigh object`33693370 Extra-high effort. Not all models accept this level.33713372 - `type: "xhigh"`33733374 - `BetaManagedAgentsEffortMax object`33753376 Maximum effort. Favors reasoning depth over latency.33773378 - `type: "max"`33793380 - `inference_geo: optional string or null`33813382 Geographic region for model inference. When unset, requests fall through to the workspace's default_inference_geo. On update, `model` is whole-object replacement — omitting inference_geo clears it.33833384 - `speed: optional "standard" or "fast" or null`33853386 Inference speed mode. `fast` provides significantly faster output token generation at premium pricing. Not all models support `fast`; invalid combinations are rejected at create time.33873388 - `"standard"`33893390 - `"fast"`33913392- `multiagent: optional BetaManagedAgentsMultiagentParams or null`33933394 A coordinator topology: the session's primary thread orchestrates work by spawning session threads, each running an agent drawn from the `agents` roster.33953396 - `type: "coordinator"`33973398 - `agents: array of BetaManagedAgentsMultiagentRosterEntryParams`33993400 Agents the coordinator may spawn as session threads. 1–20 entries. Each entry is an agent ID string, a versioned `{"type":"agent","id","version"}` reference, or `{"type":"self"}` to allow recursive self-invocation. Entries must reference distinct agents (after resolving `self` and string forms); at most one `self`. Referenced agents must exist, must not be archived, and must not themselves have `multiagent` set (depth limit 1).34013402 - `string`34033404 - `BetaManagedAgentsAgentParams object`34053406 Specification for an Agent. Provide a specific `version` or use the short-form `agent="agent_id"` for the most recent version34073408 - `type: "agent"`34093410 - `id: string`34113412 The `agent` ID.34133414 minLength: 1, maxLength: 12834153416 - `version: optional number`34173418 The specific `agent` version to use. Omit to use the latest version. Must be at least 1 if specified.34193420 format: int3234213422 - `BetaManagedAgentsMultiagentSelfParams object`34233424 Sentinel roster entry meaning "the agent that owns this configuration". Resolved server-side to a concrete agent reference.34253426 - `type: "self"`34273428 - `BetaManagedAgentsAdvisorParams object`34293430 Platform advisor roster entry: a model the session's primary thread may consult mid-turn. At most one per roster; the entry occupies the roster name `anthropic.advisor`.34313432 - `type: "advisor"`34333434 - `model: string`34353436 A Claude model id. The model must be permitted as an advisor for this agent's model — see the sessions/threads/advisor spec.34373438 minLength: 1, maxLength: 25634393440- `name: optional string`34413442 Human-readable name. Must be non-empty. Omit to preserve. Cannot be cleared.34433444 maxLength: 25634453446- `skills: optional array of BetaManagedAgentsSkillParams or null`34473448 Skills. Full replacement. Omit to preserve; send empty array or null to clear.34493450 - `BetaManagedAgentsAnthropicSkillParams object`34513452 An Anthropic-managed skill.34533454 - `type: "anthropic"`34553456 - `skill_id: string`34573458 Identifier of the Anthropic skill (e.g., "xlsx").34593460 minLength: 1, maxLength: 6434613462 - `version: optional string or null`34633464 Version to pin. Defaults to latest if omitted.34653466 minLength: 1, maxLength: 6434673468 - `BetaManagedAgentsCustomSkillParams object`34693470 A user-created custom skill.34713472 - `type: "custom"`34733474 - `skill_id: string`34753476 Tagged ID of the custom skill (e.g., "skill_01XJ5...").34773478 minLength: 1, maxLength: 6434793480 - `version: optional string or null`34813482 Version to pin. Defaults to latest if omitted.34833484 minLength: 1, maxLength: 6434853486- `system: optional string or null`34873488 System prompt. Omit to preserve; send empty string or null to clear.34893490 maxLength: 10000034913492- `tools: optional array of BetaManagedAgentsAgentToolset20260401Params or BetaManagedAgentsMCPToolsetParams or BetaManagedAgentsCustomToolParams or null`34933494 Tool configurations available to the agent. Full replacement. Omit to preserve; send empty array or null to clear. Maximum of 128 tools across all toolsets allowed.34953496 - `BetaManagedAgentsAgentToolset20260401Params object`34973498 Configuration for built-in agent tools. Use this to enable or disable groups of tools available to the agent.34993500 - `type: "agent_toolset_20260401"`35013502 - `configs: optional array of BetaManagedAgentsAgentToolConfigParams`35033504 Per-tool configuration overrides.35053506 - `BetaManagedAgentsBashToolConfigParams object`35073508 Configuration override for the bash tool.35093510 - `type: optional "bash"`35113512 - `name: "bash"`35133514 Must be "bash".35153516 - `enabled: optional boolean or null`35173518 Whether this tool is enabled and available to Claude. Overrides the default_config setting.35193520 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`35213522 Permission policy for tool execution.35233524 - `BetaManagedAgentsAlwaysAllowPolicy object`35253526 Tool calls are automatically approved without user confirmation.35273528 - `type: "always_allow"`35293530 - `BetaManagedAgentsAlwaysAskPolicy object`35313532 Tool calls require user confirmation before execution.35333534 - `type: "always_ask"`35353536 - `BetaManagedAgentsAutoPolicy object`35373538 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.35393540 - `type: "auto"`35413542 - `BetaManagedAgentsEditToolConfigParams object`35433544 Configuration override for the edit tool.35453546 - `type: optional "edit"`35473548 - `name: "edit"`35493550 Must be "edit".35513552 - `enabled: optional boolean or null`35533554 Whether this tool is enabled and available to Claude. Overrides the default_config setting.35553556 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`35573558 Permission policy for tool execution.35593560 - `BetaManagedAgentsAlwaysAllowPolicy object`35613562 Tool calls are automatically approved without user confirmation.35633564 - `BetaManagedAgentsAlwaysAskPolicy object`35653566 Tool calls require user confirmation before execution.35673568 - `BetaManagedAgentsAutoPolicy object`35693570 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.35713572 - `BetaManagedAgentsReadToolConfigParams object`35733574 Configuration override for the read tool.35753576 - `type: optional "read"`35773578 - `name: "read"`35793580 Must be "read".35813582 - `enabled: optional boolean or null`35833584 Whether this tool is enabled and available to Claude. Overrides the default_config setting.35853586 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`35873588 Permission policy for tool execution.35893590 - `BetaManagedAgentsAlwaysAllowPolicy object`35913592 Tool calls are automatically approved without user confirmation.35933594 - `BetaManagedAgentsAlwaysAskPolicy object`35953596 Tool calls require user confirmation before execution.35973598 - `BetaManagedAgentsAutoPolicy object`35993600 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.36013602 - `BetaManagedAgentsWriteToolConfigParams object`36033604 Configuration override for the write tool.36053606 - `type: optional "write"`36073608 - `name: "write"`36093610 Must be "write".36113612 - `enabled: optional boolean or null`36133614 Whether this tool is enabled and available to Claude. Overrides the default_config setting.36153616 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`36173618 Permission policy for tool execution.36193620 - `BetaManagedAgentsAlwaysAllowPolicy object`36213622 Tool calls are automatically approved without user confirmation.36233624 - `BetaManagedAgentsAlwaysAskPolicy object`36253626 Tool calls require user confirmation before execution.36273628 - `BetaManagedAgentsAutoPolicy object`36293630 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.36313632 - `BetaManagedAgentsGlobToolConfigParams object`36333634 Configuration override for the glob tool.36353636 - `type: optional "glob"`36373638 - `name: "glob"`36393640 Must be "glob".36413642 - `enabled: optional boolean or null`36433644 Whether this tool is enabled and available to Claude. Overrides the default_config setting.36453646 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`36473648 Permission policy for tool execution.36493650 - `BetaManagedAgentsAlwaysAllowPolicy object`36513652 Tool calls are automatically approved without user confirmation.36533654 - `BetaManagedAgentsAlwaysAskPolicy object`36553656 Tool calls require user confirmation before execution.36573658 - `BetaManagedAgentsAutoPolicy object`36593660 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.36613662 - `BetaManagedAgentsGrepToolConfigParams object`36633664 Configuration override for the grep tool.36653666 - `type: optional "grep"`36673668 - `name: "grep"`36693670 Must be "grep".36713672 - `enabled: optional boolean or null`36733674 Whether this tool is enabled and available to Claude. Overrides the default_config setting.36753676 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`36773678 Permission policy for tool execution.36793680 - `BetaManagedAgentsAlwaysAllowPolicy object`36813682 Tool calls are automatically approved without user confirmation.36833684 - `BetaManagedAgentsAlwaysAskPolicy object`36853686 Tool calls require user confirmation before execution.36873688 - `BetaManagedAgentsAutoPolicy object`36893690 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.36913692 - `BetaManagedAgentsWebFetchToolConfigParams object`36933694 Configuration override for the web_fetch tool.36953696 - `type: optional "web_fetch"`36973698 - `name: "web_fetch"`36993700 Must be "web_fetch".37013702 - `allowed_domains: optional array of string`37033704 Only fetch URLs whose host is one of these domains or a subdomain of one. Each entry is a plain hostname like "docs.example.com" (no scheme, port, or path). At most 64 entries; an empty list is rejected (omit the field instead). Cannot be combined with blocked_domains.37053706 - `blocked_domains: optional array of string`37073708 Never fetch URLs whose host is one of these domains or a subdomain of one. Each entry is a plain hostname like "ads.example.com" (no scheme, port, or path). At most 64 entries; an empty list is rejected (omit the field instead). Cannot be combined with allowed_domains.37093710 - `enabled: optional boolean or null`37113712 Whether this tool is enabled and available to Claude. Overrides the default_config setting.37133714 - `max_content_tokens: optional number or null`37153716 Maximum number of tokens of fetched text content to include in context per call. Does not apply to binary content such as PDFs.37173718 format: int3237193720 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`37213722 Permission policy for tool execution.37233724 - `BetaManagedAgentsAlwaysAllowPolicy object`37253726 Tool calls are automatically approved without user confirmation.37273728 - `BetaManagedAgentsAlwaysAskPolicy object`37293730 Tool calls require user confirmation before execution.37313732 - `BetaManagedAgentsAutoPolicy object`37333734 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.37353736 - `BetaManagedAgentsWebSearchToolConfigParams object`37373738 Configuration override for the web_search tool.37393740 - `type: optional "web_search"`37413742 - `name: "web_search"`37433744 Must be "web_search".37453746 - `allowed_domains: optional array of string`37473748 Only return search results whose host is one of these domains or a subdomain of one. Each entry is a plain hostname like "docs.example.com" (no scheme or port; an optional path suffix is accepted). At most 64 entries; an empty list is rejected (omit the field instead). Cannot be combined with blocked_domains.37493750 - `blocked_domains: optional array of string`37513752 Never return search results whose host is one of these domains or a subdomain of one. Each entry is a plain hostname like "ads.example.com" (no scheme or port; an optional path suffix is accepted). At most 64 entries; an empty list is rejected (omit the field instead). Cannot be combined with allowed_domains.37533754 - `enabled: optional boolean or null`37553756 Whether this tool is enabled and available to Claude. Overrides the default_config setting.37573758 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`37593760 Permission policy for tool execution.37613762 - `BetaManagedAgentsAlwaysAllowPolicy object`37633764 Tool calls are automatically approved without user confirmation.37653766 - `BetaManagedAgentsAlwaysAskPolicy object`37673768 Tool calls require user confirmation before execution.37693770 - `BetaManagedAgentsAutoPolicy object`37713772 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.37733774 - `user_location: optional BetaManagedAgentsUserLocation or null`37753776 Approximate user location for search result localization.37773778 - `type: "approximate"`37793780 Location precision. Only "approximate" is supported.37813782 - `city: optional string or null`37833784 City name.37853786 minLength: 1, maxLength: 25537873788 - `country: optional string or null`37893790 Two-letter ISO 3166-1 country code, uppercase.37913792 - `region: optional string or null`37933794 Region or state name.37953796 minLength: 1, maxLength: 25537973798 - `timezone: optional string or null`37993800 IANA timezone identifier, e.g. "America/Los_Angeles".38013802 minLength: 1, maxLength: 25538033804 - `default_config: optional BetaManagedAgentsAgentToolsetDefaultConfigParams or null`38053806 Default configuration for all tools in a toolset.38073808 - `enabled: optional boolean or null`38093810 Whether tools are enabled and available to Claude by default. Defaults to true if not specified.38113812 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`38133814 Permission policy for tool execution.38153816 - `BetaManagedAgentsAlwaysAllowPolicy object`38173818 Tool calls are automatically approved without user confirmation.38193820 - `BetaManagedAgentsAlwaysAskPolicy object`38213822 Tool calls require user confirmation before execution.38233824 - `BetaManagedAgentsAutoPolicy object`38253826 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.38273828 - `BetaManagedAgentsMCPToolsetParams object`38293830 Configuration for tools from an MCP server defined in `mcp_servers`.38313832 - `type: "mcp_toolset"`38333834 - `mcp_server_name: string`38353836 Name of the MCP server. Must match a server name from the mcp_servers array. 1-255 characters.38373838 minLength: 1, maxLength: 25538393840 - `configs: optional array of BetaManagedAgentsMCPToolConfigParams`38413842 Per-tool configuration overrides.38433844 - `name: string`38453846 Name of the MCP tool to configure. 1-128 characters.38473848 minLength: 1, maxLength: 12838493850 - `enabled: optional boolean or null`38513852 Whether this tool is enabled. Overrides the `default_config` setting.38533854 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`38553856 Permission policy for tool execution.38573858 - `BetaManagedAgentsAlwaysAllowPolicy object`38593860 Tool calls are automatically approved without user confirmation.38613862 - `BetaManagedAgentsAlwaysAskPolicy object`38633864 Tool calls require user confirmation before execution.38653866 - `BetaManagedAgentsAutoPolicy object`38673868 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.38693870 - `default_config: optional BetaManagedAgentsMCPToolsetDefaultConfigParams or null`38713872 Default configuration for all tools from an MCP server.38733874 - `enabled: optional boolean or null`38753876 Whether tools are enabled by default. Defaults to true if not specified.38773878 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`38793880 Permission policy for tool execution.38813882 - `BetaManagedAgentsAlwaysAllowPolicy object`38833884 Tool calls are automatically approved without user confirmation.38853886 - `BetaManagedAgentsAlwaysAskPolicy object`38873888 Tool calls require user confirmation before execution.38893890 - `BetaManagedAgentsAutoPolicy object`38913892 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.38933894 - `BetaManagedAgentsCustomToolParams object`38953896 A custom tool that is executed by the API client rather than the agent. When the agent calls this tool, an `agent.custom_tool_use` event is emitted and the session goes idle, waiting for the client to provide the result via a `user.custom_tool_result` event.38973898 - `type: "custom"`38993900 - `description: string`39013902 Description of what the tool does, shown to the agent to help it decide when to use the tool.39033904 minLength: 139053906 - `input_schema: BetaManagedAgentsCustomToolInputSchema`39073908 JSON Schema for custom tool input parameters.39093910 - `type: "object"`39113912 - `properties: optional map[unknown] or null`39133914 - `required: optional array of string or null`39153916 - `name: string`39173918 Unique name for the tool. 1-128 characters; letters, digits, underscores, and hyphens.39193920 minLength: 1, maxLength: 12839213922- `version: optional number`39233924 The agent's current version, used to prevent concurrent overwrites. Obtain this value from a create or retrieve response. Must be at least 1 if specified. When supplied, the request fails if it does not match the server's current version; omit to apply the update unconditionally.39253926 format: int3239273928### Returns39293930- `BetaManagedAgentsAgent object`39313932 A Managed Agents `agent`.39333934 - `type: "agent"`39353936 - `id: string`39373938 - `archived_at: string or null`39393940 A timestamp in RFC 3339 format39413942 format: date-time39433944 - `created_at: string`39453946 A timestamp in RFC 3339 format39473948 format: date-time39493950 - `description: string or null`39513952 - `mcp_servers: array of BetaManagedAgentsMCPServerURLDefinition`39533954 - `type: "url"`39553956 - `name: string`39573958 - `url: string`39593960 - `metadata: map[string]`39613962 - `model: BetaManagedAgentsModelConfig`39633964 Model identifier and configuration.39653966 - `id: BetaManagedAgentsModel`39673968 The model that will power your agent.39693970 See [models](https://docs.anthropic.com/en/docs/models-overview) for additional details and options.39713972 - `"claude-fable-5-1" or "claude-sonnet-5" or "claude-fable-5" or 11 more`39733974 The model that will power your agent.39753976 See [models](https://docs.anthropic.com/en/docs/models-overview) for additional details and options.39773978 - `"claude-fable-5-1"`39793980 Frontier intelligence for ambitious tasks across coding, scientific discovery, and enterprise workflows39813982 - `"claude-sonnet-5"`39833984 High-performance model for coding and agents39853986 - `"claude-fable-5"`39873988 Next generation of intelligence for the hardest knowledge work and coding problems39893990 - `"claude-opus-5"`39913992 Powerful intelligence for long-running agents and coding39933994 - `"claude-opus-4-8"`39953996 Powerful intelligence for long-running agents and coding39973998 - `"claude-opus-4-7"`39994000 Powerful intelligence for long-running agents and coding40014002 - `"claude-opus-4-6"`40034004 Powerful intelligence for long-running agents and coding40054006 - `"claude-sonnet-4-6"`40074008 Best combination of speed and intelligence40094010 - `"claude-haiku-4-5"`40114012 Fastest model with near-frontier intelligence40134014 - `"claude-haiku-4-5-20251001"`40154016 Fastest model with near-frontier intelligence40174018 - `"claude-opus-4-5"`40194020 Powerful intelligence for long-running agents and coding40214022 - `"claude-opus-4-5-20251101"`40234024 Powerful intelligence for long-running agents and coding40254026 - `"claude-sonnet-4-5"`40274028 High-performance model for agents and coding40294030 - `"claude-sonnet-4-5-20250929"`40314032 High-performance model for agents and coding40334034 - `string`40354036 - `effort: optional BetaManagedAgentsEffortLow or BetaManagedAgentsEffortMedium or BetaManagedAgentsEffortHigh or 2 more`40374038 How hard Claude works on each turn. Sets `output_config.effort` on every Messages call the session makes.40394040 - `BetaManagedAgentsEffortLow object`40414042 Low effort. Favors latency over reasoning depth.40434044 - `type: "low"`40454046 - `BetaManagedAgentsEffortMedium object`40474048 Medium effort. Balances latency and reasoning depth.40494050 - `type: "medium"`40514052 - `BetaManagedAgentsEffortHigh object`40534054 High effort. Favors reasoning depth.40554056 - `type: "high"`40574058 - `BetaManagedAgentsEffortXhigh object`40594060 Extra-high effort. Not all models accept this level.40614062 - `type: "xhigh"`40634064 - `BetaManagedAgentsEffortMax object`40654066 Maximum effort. Favors reasoning depth over latency.40674068 - `type: "max"`40694070 - `inference_geo: optional string`40714072 Geographic region for model inference. When unset, requests fall through to the workspace's default_inference_geo.40734074 - `speed: optional "standard" or "fast"`40754076 Inference speed mode. `fast` provides significantly faster output token generation at premium pricing. Not all models support `fast`; invalid combinations are rejected at create time.40774078 - `"standard"`40794080 - `"fast"`40814082 - `multiagent: BetaManagedAgentsMultiagent or null`40834084 Resolved coordinator topology with a concrete agent roster.40854086 - `type: "coordinator"`40874088 - `agents: array of BetaManagedAgentsAgentReference or BetaManagedAgentsAdvisor`40894090 Agents the coordinator may spawn as session threads, each resolved to a specific version.40914092 - `BetaManagedAgentsAgentReference object`40934094 A resolved agent reference with a concrete version.40954096 - `type: "agent"`40974098 - `id: string`40994100 - `version: number`41014102 format: int3241034104 - `BetaManagedAgentsAdvisor object`41054106 Platform advisor roster entry: a model the session's primary thread may consult mid-turn.41074108 - `type: "advisor"`41094110 - `model: string`41114112 The advisor model id.41134114 - `name: string`41154116 - `skills: array of BetaManagedAgentsAnthropicSkill or BetaManagedAgentsCustomSkill`41174118 - `BetaManagedAgentsAnthropicSkill object`41194120 A resolved Anthropic-managed skill.41214122 - `type: "anthropic"`41234124 - `skill_id: string`41254126 - `version: string`41274128 - `BetaManagedAgentsCustomSkill object`41294130 A resolved user-created custom skill.41314132 - `type: "custom"`41334134 - `skill_id: string`41354136 - `version: string`41374138 - `system: string or null`41394140 - `tools: array of BetaManagedAgentsAgentToolset20260401 or BetaManagedAgentsMCPToolset or BetaManagedAgentsCustomTool`41414142 - `BetaManagedAgentsAgentToolset20260401 object`41434144 - `type: "agent_toolset_20260401"`41454146 - `configs: array of BetaManagedAgentsAgentToolConfig`41474148 - `BetaManagedAgentsBashToolConfig object`41494150 Configuration for the bash tool.41514152 - `type: "bash"`41534154 - `enabled: boolean`41554156 - `name: "bash"`41574158 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`41594160 Permission policy for tool execution.41614162 - `BetaManagedAgentsAlwaysAllowPolicy object`41634164 Tool calls are automatically approved without user confirmation.41654166 - `type: "always_allow"`41674168 - `BetaManagedAgentsAlwaysAskPolicy object`41694170 Tool calls require user confirmation before execution.41714172 - `type: "always_ask"`41734174 - `BetaManagedAgentsAutoPolicy object`41754176 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.41774178 - `type: "auto"`41794180 - `BetaManagedAgentsEditToolConfig object`41814182 Configuration for the edit tool.41834184 - `type: "edit"`41854186 - `enabled: boolean`41874188 - `name: "edit"`41894190 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`41914192 Permission policy for tool execution.41934194 - `BetaManagedAgentsAlwaysAllowPolicy object`41954196 Tool calls are automatically approved without user confirmation.41974198 - `BetaManagedAgentsAlwaysAskPolicy object`41994200 Tool calls require user confirmation before execution.42014202 - `BetaManagedAgentsAutoPolicy object`42034204 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.42054206 - `BetaManagedAgentsReadToolConfig object`42074208 Configuration for the read tool.42094210 - `type: "read"`42114212 - `enabled: boolean`42134214 - `name: "read"`42154216 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`42174218 Permission policy for tool execution.42194220 - `BetaManagedAgentsAlwaysAllowPolicy object`42214222 Tool calls are automatically approved without user confirmation.42234224 - `BetaManagedAgentsAlwaysAskPolicy object`42254226 Tool calls require user confirmation before execution.42274228 - `BetaManagedAgentsAutoPolicy object`42294230 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.42314232 - `BetaManagedAgentsWriteToolConfig object`42334234 Configuration for the write tool.42354236 - `type: "write"`42374238 - `enabled: boolean`42394240 - `name: "write"`42414242 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`42434244 Permission policy for tool execution.42454246 - `BetaManagedAgentsAlwaysAllowPolicy object`42474248 Tool calls are automatically approved without user confirmation.42494250 - `BetaManagedAgentsAlwaysAskPolicy object`42514252 Tool calls require user confirmation before execution.42534254 - `BetaManagedAgentsAutoPolicy object`42554256 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.42574258 - `BetaManagedAgentsGlobToolConfig object`42594260 Configuration for the glob tool.42614262 - `type: "glob"`42634264 - `enabled: boolean`42654266 - `name: "glob"`42674268 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`42694270 Permission policy for tool execution.42714272 - `BetaManagedAgentsAlwaysAllowPolicy object`42734274 Tool calls are automatically approved without user confirmation.42754276 - `BetaManagedAgentsAlwaysAskPolicy object`42774278 Tool calls require user confirmation before execution.42794280 - `BetaManagedAgentsAutoPolicy object`42814282 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.42834284 - `BetaManagedAgentsGrepToolConfig object`42854286 Configuration for the grep tool.42874288 - `type: "grep"`42894290 - `enabled: boolean`42914292 - `name: "grep"`42934294 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`42954296 Permission policy for tool execution.42974298 - `BetaManagedAgentsAlwaysAllowPolicy object`42994300 Tool calls are automatically approved without user confirmation.43014302 - `BetaManagedAgentsAlwaysAskPolicy object`43034304 Tool calls require user confirmation before execution.43054306 - `BetaManagedAgentsAutoPolicy object`43074308 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.43094310 - `BetaManagedAgentsWebFetchToolConfig object`43114312 Configuration for the web_fetch tool.43134314 - `type: "web_fetch"`43154316 - `enabled: boolean`43174318 - `name: "web_fetch"`43194320 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`43214322 Permission policy for tool execution.43234324 - `BetaManagedAgentsAlwaysAllowPolicy object`43254326 Tool calls are automatically approved without user confirmation.43274328 - `BetaManagedAgentsAlwaysAskPolicy object`43294330 Tool calls require user confirmation before execution.43314332 - `BetaManagedAgentsAutoPolicy object`43334334 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.43354336 - `allowed_domains: optional array of string`43374338 - `blocked_domains: optional array of string`43394340 - `max_content_tokens: optional number or null`43414342 format: int3243434344 - `BetaManagedAgentsWebSearchToolConfig object`43454346 Configuration for the web_search tool.43474348 - `type: "web_search"`43494350 - `enabled: boolean`43514352 - `name: "web_search"`43534354 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`43554356 Permission policy for tool execution.43574358 - `BetaManagedAgentsAlwaysAllowPolicy object`43594360 Tool calls are automatically approved without user confirmation.43614362 - `BetaManagedAgentsAlwaysAskPolicy object`43634364 Tool calls require user confirmation before execution.43654366 - `BetaManagedAgentsAutoPolicy object`43674368 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.43694370 - `allowed_domains: optional array of string`43714372 - `blocked_domains: optional array of string`43734374 - `user_location: optional BetaManagedAgentsUserLocation or null`43754376 Approximate user location for search result localization.43774378 - `type: "approximate"`43794380 Location precision. Only "approximate" is supported.43814382 - `city: optional string or null`43834384 City name.43854386 minLength: 1, maxLength: 25543874388 - `country: optional string or null`43894390 Two-letter ISO 3166-1 country code, uppercase.43914392 - `region: optional string or null`43934394 Region or state name.43954396 minLength: 1, maxLength: 25543974398 - `timezone: optional string or null`43994400 IANA timezone identifier, e.g. "America/Los_Angeles".44014402 minLength: 1, maxLength: 25544034404 - `default_config: BetaManagedAgentsAgentToolsetDefaultConfig`44054406 Resolved default configuration for agent tools.44074408 - `enabled: boolean`44094410 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`44114412 Permission policy for tool execution.44134414 - `BetaManagedAgentsAlwaysAllowPolicy object`44154416 Tool calls are automatically approved without user confirmation.44174418 - `BetaManagedAgentsAlwaysAskPolicy object`44194420 Tool calls require user confirmation before execution.44214422 - `BetaManagedAgentsAutoPolicy object`44234424 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.44254426 - `BetaManagedAgentsMCPToolset object`44274428 - `type: "mcp_toolset"`44294430 - `configs: array of BetaManagedAgentsMCPToolConfig`44314432 - `enabled: boolean`44334434 - `name: string`44354436 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`44374438 Permission policy for tool execution.44394440 - `BetaManagedAgentsAlwaysAllowPolicy object`44414442 Tool calls are automatically approved without user confirmation.44434444 - `BetaManagedAgentsAlwaysAskPolicy object`44454446 Tool calls require user confirmation before execution.44474448 - `BetaManagedAgentsAutoPolicy object`44494450 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.44514452 - `default_config: BetaManagedAgentsMCPToolsetDefaultConfig`44534454 Resolved default configuration for all tools from an MCP server.44554456 - `enabled: boolean`44574458 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`44594460 Permission policy for tool execution.44614462 - `BetaManagedAgentsAlwaysAllowPolicy object`44634464 Tool calls are automatically approved without user confirmation.44654466 - `BetaManagedAgentsAlwaysAskPolicy object`44674468 Tool calls require user confirmation before execution.44694470 - `BetaManagedAgentsAutoPolicy object`44714472 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.44734474 - `mcp_server_name: string`44754476 - `BetaManagedAgentsCustomTool object`44774478 A custom tool as returned in API responses.44794480 - `type: "custom"`44814482 - `description: string`44834484 - `input_schema: BetaManagedAgentsCustomToolInputSchema`44854486 JSON Schema for custom tool input parameters.44874488 - `type: "object"`44894490 - `properties: optional map[unknown] or null`44914492 - `required: optional array of string or null`44934494 - `name: string`44954496 - `updated_at: string`44974498 A timestamp in RFC 3339 format44994500 format: date-time45014502 - `version: number`45034504 The agent's current version. Starts at 1 and increments when the agent is modified.45054506 format: int3245074508### Example45094510```bash4511curl https://api.anthropic.com/v1/agents/$AGENT_ID \4512 -H 'Content-Type: application/json' \4513 -H 'anthropic-version: 2023-06-01' \4514 -H 'anthropic-beta: managed-agents-2026-04-01' \4515 -H "X-Api-Key: $ANTHROPIC_API_KEY" \4516 -d '{4517 "description": "updated",4518 "system": "You are a general-purpose agent that can research, write code, run commands, and use connected tools to complete the user'\''s task end to end.",4519 "version": 14520 }'4521```45224523#### Response (200)45244525```json4526{4527 "id": "agent_011CZkYpogX7uDKUyvBTophP",4528 "archived_at": null,4529 "created_at": "2026-03-15T10:00:00Z",4530 "description": "A general-purpose starter agent.",4531 "mcp_servers": [4532 {4533 "name": "example-mcp",4534 "type": "url",4535 "url": "https://example-server.modelcontextprotocol.io/sse"4536 }4537 ],4538 "metadata": {4539 "foo": "bar"4540 },4541 "model": {4542 "id": "claude-opus-5",4543 "effort": {4544 "type": "low"4545 },4546 "inference_geo": "inference_geo",4547 "speed": "standard"4548 },4549 "multiagent": {4550 "agents": [4551 {4552 "id": "agent_011CZkYqphY8vELVzwCUpqiQ",4553 "type": "agent",4554 "version": 14555 }4556 ],4557 "type": "coordinator"4558 },4559 "name": "My First Agent",4560 "skills": [4561 {4562 "skill_id": "xlsx",4563 "type": "anthropic",4564 "version": "1"4565 },4566 {4567 "skill_id": "skill_011CZkZFNu9hAbo3jZPRgTlx",4568 "type": "custom",4569 "version": "2"4570 }4571 ],4572 "system": "You are a general-purpose agent that can research, write code, run commands, and use connected tools to complete the user's task end to end.",4573 "tools": [4574 {4575 "configs": [4576 {4577 "enabled": true,4578 "name": "bash",4579 "permission_policy": {4580 "type": "always_allow"4581 },4582 "type": "bash"4583 }4584 ],4585 "default_config": {4586 "enabled": true,4587 "permission_policy": {4588 "type": "always_ask"4589 }4590 },4591 "type": "agent_toolset_20260401"4592 }4593 ],4594 "type": "agent",4595 "updated_at": "2026-03-15T10:00:00Z",4596 "version": 14597}4598```45994600## Archive Agent46014602**POST** `/v1/agents/{agent_id}/archive`46034604Archive Agent46054606### Path parameters46074608- `agent_id: string`46094610### Headers46114612- `"anthropic-beta": optional array of AnthropicBeta`46134614 Optional header to specify the beta version(s) you want to use.46154616 - `string`46174618 - `"message-batches-2024-09-24" or "prompt-caching-2024-07-31" or "computer-use-2024-10-22" or 43 more`46194620 - `"message-batches-2024-09-24"`46214622 - `"prompt-caching-2024-07-31"`46234624 - `"computer-use-2024-10-22"`46254626 - `"computer-use-2025-01-24"`46274628 - `"pdfs-2024-09-25"`46294630 - `"token-counting-2024-11-01"`46314632 - `"token-efficient-tools-2025-02-19"`46334634 - `"output-128k-2025-02-19"`46354636 - `"files-api-2025-04-14"`46374638 - `"mcp-client-2025-04-04"`46394640 - `"mcp-client-2025-11-20"`46414642 - `"dev-full-thinking-2025-05-14"`46434644 - `"interleaved-thinking-2025-05-14"`46454646 - `"code-execution-2025-05-22"`46474648 - `"extended-cache-ttl-2025-04-11"`46494650 - `"context-1m-2025-08-07"`46514652 - `"context-management-2025-06-27"`46534654 - `"model-context-window-exceeded-2025-08-26"`46554656 - `"skills-2025-10-02"`46574658 - `"fast-mode-2026-02-01"`46594660 - `"output-300k-2026-03-24"`46614662 - `"user-profiles-2026-03-24"`46634664 - `"user-profiles-2026-08-18"`46654666 - `"user-profiles-2026-09-04"`46674668 - `"advisor-tool-2026-03-01"`46694670 - `"managed-agents-2026-04-01"`46714672 - `"cache-diagnosis-2026-04-07"`46734674 - `"dreaming-2026-04-21"`46754676 - `"thinking-token-count-2026-05-13"`46774678 - `"server-side-fallback-2026-06-01"`46794680 - `"server-side-fallback-2026-07-01"`46814682 - `"fallback-credit-2026-06-01"`46834684 - `"fallback-credit-2026-07-01"`46854686 - `"agent-memory-2026-07-22"`46874688 - `"mid-conversation-tool-changes-2026-07-01"`46894690 - `"compact-2026-01-12"`46914692 - `"computer-use-2025-11-24"`46934694 - `"mcp-tunnels-2026-06-22"`46954696 - `"structured-outputs-2025-11-13"`46974698 - `"task-budgets-2026-03-13"`46994700 - `"thinking-display-updates-2026-08-18"`47014702 - `"ce-user-management-2026-07-13"`47034704 - `"mid-conversation-output-config-2026-07-01"`47054706 - `"thinking-binding-controls-2026-08-01"`47074708 - `"mid-conversation-system-clear-at-2026-08-21"`47094710 - `"compact-2026-09-04"`47114712- `"anthropic-workspace-id": optional string`47134714### Returns47154716- `BetaManagedAgentsAgent object`47174718 A Managed Agents `agent`.47194720 - `type: "agent"`47214722 - `id: string`47234724 - `archived_at: string or null`47254726 A timestamp in RFC 3339 format47274728 format: date-time47294730 - `created_at: string`47314732 A timestamp in RFC 3339 format47334734 format: date-time47354736 - `description: string or null`47374738 - `mcp_servers: array of BetaManagedAgentsMCPServerURLDefinition`47394740 - `type: "url"`47414742 - `name: string`47434744 - `url: string`47454746 - `metadata: map[string]`47474748 - `model: BetaManagedAgentsModelConfig`47494750 Model identifier and configuration.47514752 - `id: BetaManagedAgentsModel`47534754 The model that will power your agent.47554756 See [models](https://docs.anthropic.com/en/docs/models-overview) for additional details and options.47574758 - `"claude-fable-5-1" or "claude-sonnet-5" or "claude-fable-5" or 11 more`47594760 The model that will power your agent.47614762 See [models](https://docs.anthropic.com/en/docs/models-overview) for additional details and options.47634764 - `"claude-fable-5-1"`47654766 Frontier intelligence for ambitious tasks across coding, scientific discovery, and enterprise workflows47674768 - `"claude-sonnet-5"`47694770 High-performance model for coding and agents47714772 - `"claude-fable-5"`47734774 Next generation of intelligence for the hardest knowledge work and coding problems47754776 - `"claude-opus-5"`47774778 Powerful intelligence for long-running agents and coding47794780 - `"claude-opus-4-8"`47814782 Powerful intelligence for long-running agents and coding47834784 - `"claude-opus-4-7"`47854786 Powerful intelligence for long-running agents and coding47874788 - `"claude-opus-4-6"`47894790 Powerful intelligence for long-running agents and coding47914792 - `"claude-sonnet-4-6"`47934794 Best combination of speed and intelligence47954796 - `"claude-haiku-4-5"`47974798 Fastest model with near-frontier intelligence47994800 - `"claude-haiku-4-5-20251001"`48014802 Fastest model with near-frontier intelligence48034804 - `"claude-opus-4-5"`48054806 Powerful intelligence for long-running agents and coding48074808 - `"claude-opus-4-5-20251101"`48094810 Powerful intelligence for long-running agents and coding48114812 - `"claude-sonnet-4-5"`48134814 High-performance model for agents and coding48154816 - `"claude-sonnet-4-5-20250929"`48174818 High-performance model for agents and coding48194820 - `string`48214822 - `effort: optional BetaManagedAgentsEffortLow or BetaManagedAgentsEffortMedium or BetaManagedAgentsEffortHigh or 2 more`48234824 How hard Claude works on each turn. Sets `output_config.effort` on every Messages call the session makes.48254826 - `BetaManagedAgentsEffortLow object`48274828 Low effort. Favors latency over reasoning depth.48294830 - `type: "low"`48314832 - `BetaManagedAgentsEffortMedium object`48334834 Medium effort. Balances latency and reasoning depth.48354836 - `type: "medium"`48374838 - `BetaManagedAgentsEffortHigh object`48394840 High effort. Favors reasoning depth.48414842 - `type: "high"`48434844 - `BetaManagedAgentsEffortXhigh object`48454846 Extra-high effort. Not all models accept this level.48474848 - `type: "xhigh"`48494850 - `BetaManagedAgentsEffortMax object`48514852 Maximum effort. Favors reasoning depth over latency.48534854 - `type: "max"`48554856 - `inference_geo: optional string`48574858 Geographic region for model inference. When unset, requests fall through to the workspace's default_inference_geo.48594860 - `speed: optional "standard" or "fast"`48614862 Inference speed mode. `fast` provides significantly faster output token generation at premium pricing. Not all models support `fast`; invalid combinations are rejected at create time.48634864 - `"standard"`48654866 - `"fast"`48674868 - `multiagent: BetaManagedAgentsMultiagent or null`48694870 Resolved coordinator topology with a concrete agent roster.48714872 - `type: "coordinator"`48734874 - `agents: array of BetaManagedAgentsAgentReference or BetaManagedAgentsAdvisor`48754876 Agents the coordinator may spawn as session threads, each resolved to a specific version.48774878 - `BetaManagedAgentsAgentReference object`48794880 A resolved agent reference with a concrete version.48814882 - `type: "agent"`48834884 - `id: string`48854886 - `version: number`48874888 format: int3248894890 - `BetaManagedAgentsAdvisor object`48914892 Platform advisor roster entry: a model the session's primary thread may consult mid-turn.48934894 - `type: "advisor"`48954896 - `model: string`48974898 The advisor model id.48994900 - `name: string`49014902 - `skills: array of BetaManagedAgentsAnthropicSkill or BetaManagedAgentsCustomSkill`49034904 - `BetaManagedAgentsAnthropicSkill object`49054906 A resolved Anthropic-managed skill.49074908 - `type: "anthropic"`49094910 - `skill_id: string`49114912 - `version: string`49134914 - `BetaManagedAgentsCustomSkill object`49154916 A resolved user-created custom skill.49174918 - `type: "custom"`49194920 - `skill_id: string`49214922 - `version: string`49234924 - `system: string or null`49254926 - `tools: array of BetaManagedAgentsAgentToolset20260401 or BetaManagedAgentsMCPToolset or BetaManagedAgentsCustomTool`49274928 - `BetaManagedAgentsAgentToolset20260401 object`49294930 - `type: "agent_toolset_20260401"`49314932 - `configs: array of BetaManagedAgentsAgentToolConfig`49334934 - `BetaManagedAgentsBashToolConfig object`49354936 Configuration for the bash tool.49374938 - `type: "bash"`49394940 - `enabled: boolean`49414942 - `name: "bash"`49434944 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`49454946 Permission policy for tool execution.49474948 - `BetaManagedAgentsAlwaysAllowPolicy object`49494950 Tool calls are automatically approved without user confirmation.49514952 - `type: "always_allow"`49534954 - `BetaManagedAgentsAlwaysAskPolicy object`49554956 Tool calls require user confirmation before execution.49574958 - `type: "always_ask"`49594960 - `BetaManagedAgentsAutoPolicy object`49614962 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.49634964 - `type: "auto"`49654966 - `BetaManagedAgentsEditToolConfig object`49674968 Configuration for the edit tool.49694970 - `type: "edit"`49714972 - `enabled: boolean`49734974 - `name: "edit"`49754976 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`49774978 Permission policy for tool execution.49794980 - `BetaManagedAgentsAlwaysAllowPolicy object`49814982 Tool calls are automatically approved without user confirmation.49834984 - `BetaManagedAgentsAlwaysAskPolicy object`49854986 Tool calls require user confirmation before execution.49874988 - `BetaManagedAgentsAutoPolicy object`49894990 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.49914992 - `BetaManagedAgentsReadToolConfig object`49934994 Configuration for the read tool.49954996 - `type: "read"`49974998 - `enabled: boolean`49995000 - `name: "read"`50015002 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`50035004 Permission policy for tool execution.50055006 - `BetaManagedAgentsAlwaysAllowPolicy object`50075008 Tool calls are automatically approved without user confirmation.50095010 - `BetaManagedAgentsAlwaysAskPolicy object`50115012 Tool calls require user confirmation before execution.50135014 - `BetaManagedAgentsAutoPolicy object`50155016 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.50175018 - `BetaManagedAgentsWriteToolConfig object`50195020 Configuration for the write tool.50215022 - `type: "write"`50235024 - `enabled: boolean`50255026 - `name: "write"`50275028 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`50295030 Permission policy for tool execution.50315032 - `BetaManagedAgentsAlwaysAllowPolicy object`50335034 Tool calls are automatically approved without user confirmation.50355036 - `BetaManagedAgentsAlwaysAskPolicy object`50375038 Tool calls require user confirmation before execution.50395040 - `BetaManagedAgentsAutoPolicy object`50415042 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.50435044 - `BetaManagedAgentsGlobToolConfig object`50455046 Configuration for the glob tool.50475048 - `type: "glob"`50495050 - `enabled: boolean`50515052 - `name: "glob"`50535054 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`50555056 Permission policy for tool execution.50575058 - `BetaManagedAgentsAlwaysAllowPolicy object`50595060 Tool calls are automatically approved without user confirmation.50615062 - `BetaManagedAgentsAlwaysAskPolicy object`50635064 Tool calls require user confirmation before execution.50655066 - `BetaManagedAgentsAutoPolicy object`50675068 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.50695070 - `BetaManagedAgentsGrepToolConfig object`50715072 Configuration for the grep tool.50735074 - `type: "grep"`50755076 - `enabled: boolean`50775078 - `name: "grep"`50795080 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`50815082 Permission policy for tool execution.50835084 - `BetaManagedAgentsAlwaysAllowPolicy object`50855086 Tool calls are automatically approved without user confirmation.50875088 - `BetaManagedAgentsAlwaysAskPolicy object`50895090 Tool calls require user confirmation before execution.50915092 - `BetaManagedAgentsAutoPolicy object`50935094 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.50955096 - `BetaManagedAgentsWebFetchToolConfig object`50975098 Configuration for the web_fetch tool.50995100 - `type: "web_fetch"`51015102 - `enabled: boolean`51035104 - `name: "web_fetch"`51055106 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`51075108 Permission policy for tool execution.51095110 - `BetaManagedAgentsAlwaysAllowPolicy object`51115112 Tool calls are automatically approved without user confirmation.51135114 - `BetaManagedAgentsAlwaysAskPolicy object`51155116 Tool calls require user confirmation before execution.51175118 - `BetaManagedAgentsAutoPolicy object`51195120 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.51215122 - `allowed_domains: optional array of string`51235124 - `blocked_domains: optional array of string`51255126 - `max_content_tokens: optional number or null`51275128 format: int3251295130 - `BetaManagedAgentsWebSearchToolConfig object`51315132 Configuration for the web_search tool.51335134 - `type: "web_search"`51355136 - `enabled: boolean`51375138 - `name: "web_search"`51395140 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`51415142 Permission policy for tool execution.51435144 - `BetaManagedAgentsAlwaysAllowPolicy object`51455146 Tool calls are automatically approved without user confirmation.51475148 - `BetaManagedAgentsAlwaysAskPolicy object`51495150 Tool calls require user confirmation before execution.51515152 - `BetaManagedAgentsAutoPolicy object`51535154 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.51555156 - `allowed_domains: optional array of string`51575158 - `blocked_domains: optional array of string`51595160 - `user_location: optional BetaManagedAgentsUserLocation or null`51615162 Approximate user location for search result localization.51635164 - `type: "approximate"`51655166 Location precision. Only "approximate" is supported.51675168 - `city: optional string or null`51695170 City name.51715172 minLength: 1, maxLength: 25551735174 - `country: optional string or null`51755176 Two-letter ISO 3166-1 country code, uppercase.51775178 - `region: optional string or null`51795180 Region or state name.51815182 minLength: 1, maxLength: 25551835184 - `timezone: optional string or null`51855186 IANA timezone identifier, e.g. "America/Los_Angeles".51875188 minLength: 1, maxLength: 25551895190 - `default_config: BetaManagedAgentsAgentToolsetDefaultConfig`51915192 Resolved default configuration for agent tools.51935194 - `enabled: boolean`51955196 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`51975198 Permission policy for tool execution.51995200 - `BetaManagedAgentsAlwaysAllowPolicy object`52015202 Tool calls are automatically approved without user confirmation.52035204 - `BetaManagedAgentsAlwaysAskPolicy object`52055206 Tool calls require user confirmation before execution.52075208 - `BetaManagedAgentsAutoPolicy object`52095210 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.52115212 - `BetaManagedAgentsMCPToolset object`52135214 - `type: "mcp_toolset"`52155216 - `configs: array of BetaManagedAgentsMCPToolConfig`52175218 - `enabled: boolean`52195220 - `name: string`52215222 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`52235224 Permission policy for tool execution.52255226 - `BetaManagedAgentsAlwaysAllowPolicy object`52275228 Tool calls are automatically approved without user confirmation.52295230 - `BetaManagedAgentsAlwaysAskPolicy object`52315232 Tool calls require user confirmation before execution.52335234 - `BetaManagedAgentsAutoPolicy object`52355236 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.52375238 - `default_config: BetaManagedAgentsMCPToolsetDefaultConfig`52395240 Resolved default configuration for all tools from an MCP server.52415242 - `enabled: boolean`52435244 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`52455246 Permission policy for tool execution.52475248 - `BetaManagedAgentsAlwaysAllowPolicy object`52495250 Tool calls are automatically approved without user confirmation.52515252 - `BetaManagedAgentsAlwaysAskPolicy object`52535254 Tool calls require user confirmation before execution.52555256 - `BetaManagedAgentsAutoPolicy object`52575258 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.52595260 - `mcp_server_name: string`52615262 - `BetaManagedAgentsCustomTool object`52635264 A custom tool as returned in API responses.52655266 - `type: "custom"`52675268 - `description: string`52695270 - `input_schema: BetaManagedAgentsCustomToolInputSchema`52715272 JSON Schema for custom tool input parameters.52735274 - `type: "object"`52755276 - `properties: optional map[unknown] or null`52775278 - `required: optional array of string or null`52795280 - `name: string`52815282 - `updated_at: string`52835284 A timestamp in RFC 3339 format52855286 format: date-time52875288 - `version: number`52895290 The agent's current version. Starts at 1 and increments when the agent is modified.52915292 format: int3252935294### Example52955296```bash5297curl https://api.anthropic.com/v1/agents/$AGENT_ID/archive \5298 -X POST \5299 -H 'anthropic-version: 2023-06-01' \5300 -H 'anthropic-beta: managed-agents-2026-04-01' \5301 -H "X-Api-Key: $ANTHROPIC_API_KEY"5302```53035304#### Response (200)53055306```json5307{5308 "id": "agent_011CZkYpogX7uDKUyvBTophP",5309 "archived_at": null,5310 "created_at": "2026-03-15T10:00:00Z",5311 "description": "A general-purpose starter agent.",5312 "mcp_servers": [5313 {5314 "name": "example-mcp",5315 "type": "url",5316 "url": "https://example-server.modelcontextprotocol.io/sse"5317 }5318 ],5319 "metadata": {5320 "foo": "bar"5321 },5322 "model": {5323 "id": "claude-opus-5",5324 "effort": {5325 "type": "low"5326 },5327 "inference_geo": "inference_geo",5328 "speed": "standard"5329 },5330 "multiagent": {5331 "agents": [5332 {5333 "id": "agent_011CZkYqphY8vELVzwCUpqiQ",5334 "type": "agent",5335 "version": 15336 }5337 ],5338 "type": "coordinator"5339 },5340 "name": "My First Agent",5341 "skills": [5342 {5343 "skill_id": "xlsx",5344 "type": "anthropic",5345 "version": "1"5346 },5347 {5348 "skill_id": "skill_011CZkZFNu9hAbo3jZPRgTlx",5349 "type": "custom",5350 "version": "2"5351 }5352 ],5353 "system": "You are a general-purpose agent that can research, write code, run commands, and use connected tools to complete the user's task end to end.",5354 "tools": [5355 {5356 "configs": [5357 {5358 "enabled": true,5359 "name": "bash",5360 "permission_policy": {5361 "type": "always_allow"5362 },5363 "type": "bash"5364 }5365 ],5366 "default_config": {5367 "enabled": true,5368 "permission_policy": {5369 "type": "always_ask"5370 }5371 },5372 "type": "agent_toolset_20260401"5373 }5374 ],5375 "type": "agent",5376 "updated_at": "2026-03-15T10:00:00Z",5377 "version": 15378}5379```53805381## Domain types53825383### Beta Managed Agents Advisor53845385- `BetaManagedAgentsAdvisor object`53865387 Platform advisor roster entry: a model the session's primary thread may consult mid-turn.53885389 - `type: "advisor"`53905391 - `model: string`53925393 The advisor model id.53945395### Beta Managed Agents Agent53965397- `BetaManagedAgentsAgent object`53985399 A Managed Agents `agent`.54005401 - `type: "agent"`54025403 - `id: string`54045405 - `archived_at: string or null`54065407 A timestamp in RFC 3339 format54085409 format: date-time54105411 - `created_at: string`54125413 A timestamp in RFC 3339 format54145415 format: date-time54165417 - `description: string or null`54185419 - `mcp_servers: array of BetaManagedAgentsMCPServerURLDefinition`54205421 - `type: "url"`54225423 - `name: string`54245425 - `url: string`54265427 - `metadata: map[string]`54285429 - `model: BetaManagedAgentsModelConfig`54305431 Model identifier and configuration.54325433 - `id: BetaManagedAgentsModel`54345435 The model that will power your agent.54365437 See [models](https://docs.anthropic.com/en/docs/models-overview) for additional details and options.54385439 - `"claude-fable-5-1" or "claude-sonnet-5" or "claude-fable-5" or 11 more`54405441 The model that will power your agent.54425443 See [models](https://docs.anthropic.com/en/docs/models-overview) for additional details and options.54445445 - `"claude-fable-5-1"`54465447 Frontier intelligence for ambitious tasks across coding, scientific discovery, and enterprise workflows54485449 - `"claude-sonnet-5"`54505451 High-performance model for coding and agents54525453 - `"claude-fable-5"`54545455 Next generation of intelligence for the hardest knowledge work and coding problems54565457 - `"claude-opus-5"`54585459 Powerful intelligence for long-running agents and coding54605461 - `"claude-opus-4-8"`54625463 Powerful intelligence for long-running agents and coding54645465 - `"claude-opus-4-7"`54665467 Powerful intelligence for long-running agents and coding54685469 - `"claude-opus-4-6"`54705471 Powerful intelligence for long-running agents and coding54725473 - `"claude-sonnet-4-6"`54745475 Best combination of speed and intelligence54765477 - `"claude-haiku-4-5"`54785479 Fastest model with near-frontier intelligence54805481 - `"claude-haiku-4-5-20251001"`54825483 Fastest model with near-frontier intelligence54845485 - `"claude-opus-4-5"`54865487 Powerful intelligence for long-running agents and coding54885489 - `"claude-opus-4-5-20251101"`54905491 Powerful intelligence for long-running agents and coding54925493 - `"claude-sonnet-4-5"`54945495 High-performance model for agents and coding54965497 - `"claude-sonnet-4-5-20250929"`54985499 High-performance model for agents and coding55005501 - `string`55025503 - `effort: optional BetaManagedAgentsEffortLow or BetaManagedAgentsEffortMedium or BetaManagedAgentsEffortHigh or 2 more`55045505 How hard Claude works on each turn. Sets `output_config.effort` on every Messages call the session makes.55065507 - `BetaManagedAgentsEffortLow object`55085509 Low effort. Favors latency over reasoning depth.55105511 - `type: "low"`55125513 - `BetaManagedAgentsEffortMedium object`55145515 Medium effort. Balances latency and reasoning depth.55165517 - `type: "medium"`55185519 - `BetaManagedAgentsEffortHigh object`55205521 High effort. Favors reasoning depth.55225523 - `type: "high"`55245525 - `BetaManagedAgentsEffortXhigh object`55265527 Extra-high effort. Not all models accept this level.55285529 - `type: "xhigh"`55305531 - `BetaManagedAgentsEffortMax object`55325533 Maximum effort. Favors reasoning depth over latency.55345535 - `type: "max"`55365537 - `inference_geo: optional string`55385539 Geographic region for model inference. When unset, requests fall through to the workspace's default_inference_geo.55405541 - `speed: optional "standard" or "fast"`55425543 Inference speed mode. `fast` provides significantly faster output token generation at premium pricing. Not all models support `fast`; invalid combinations are rejected at create time.55445545 - `"standard"`55465547 - `"fast"`55485549 - `multiagent: BetaManagedAgentsMultiagent or null`55505551 Resolved coordinator topology with a concrete agent roster.55525553 - `type: "coordinator"`55545555 - `agents: array of BetaManagedAgentsAgentReference or BetaManagedAgentsAdvisor`55565557 Agents the coordinator may spawn as session threads, each resolved to a specific version.55585559 - `BetaManagedAgentsAgentReference object`55605561 A resolved agent reference with a concrete version.55625563 - `type: "agent"`55645565 - `id: string`55665567 - `version: number`55685569 format: int3255705571 - `BetaManagedAgentsAdvisor object`55725573 Platform advisor roster entry: a model the session's primary thread may consult mid-turn.55745575 - `type: "advisor"`55765577 - `model: string`55785579 The advisor model id.55805581 - `name: string`55825583 - `skills: array of BetaManagedAgentsAnthropicSkill or BetaManagedAgentsCustomSkill`55845585 - `BetaManagedAgentsAnthropicSkill object`55865587 A resolved Anthropic-managed skill.55885589 - `type: "anthropic"`55905591 - `skill_id: string`55925593 - `version: string`55945595 - `BetaManagedAgentsCustomSkill object`55965597 A resolved user-created custom skill.55985599 - `type: "custom"`56005601 - `skill_id: string`56025603 - `version: string`56045605 - `system: string or null`56065607 - `tools: array of BetaManagedAgentsAgentToolset20260401 or BetaManagedAgentsMCPToolset or BetaManagedAgentsCustomTool`56085609 - `BetaManagedAgentsAgentToolset20260401 object`56105611 - `type: "agent_toolset_20260401"`56125613 - `configs: array of BetaManagedAgentsAgentToolConfig`56145615 - `BetaManagedAgentsBashToolConfig object`56165617 Configuration for the bash tool.56185619 - `type: "bash"`56205621 - `enabled: boolean`56225623 - `name: "bash"`56245625 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`56265627 Permission policy for tool execution.56285629 - `BetaManagedAgentsAlwaysAllowPolicy object`56305631 Tool calls are automatically approved without user confirmation.56325633 - `type: "always_allow"`56345635 - `BetaManagedAgentsAlwaysAskPolicy object`56365637 Tool calls require user confirmation before execution.56385639 - `type: "always_ask"`56405641 - `BetaManagedAgentsAutoPolicy object`56425643 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.56445645 - `type: "auto"`56465647 - `BetaManagedAgentsEditToolConfig object`56485649 Configuration for the edit tool.56505651 - `type: "edit"`56525653 - `enabled: boolean`56545655 - `name: "edit"`56565657 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`56585659 Permission policy for tool execution.56605661 - `BetaManagedAgentsAlwaysAllowPolicy object`56625663 Tool calls are automatically approved without user confirmation.56645665 - `BetaManagedAgentsAlwaysAskPolicy object`56665667 Tool calls require user confirmation before execution.56685669 - `BetaManagedAgentsAutoPolicy object`56705671 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.56725673 - `BetaManagedAgentsReadToolConfig object`56745675 Configuration for the read tool.56765677 - `type: "read"`56785679 - `enabled: boolean`56805681 - `name: "read"`56825683 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`56845685 Permission policy for tool execution.56865687 - `BetaManagedAgentsAlwaysAllowPolicy object`56885689 Tool calls are automatically approved without user confirmation.56905691 - `BetaManagedAgentsAlwaysAskPolicy object`56925693 Tool calls require user confirmation before execution.56945695 - `BetaManagedAgentsAutoPolicy object`56965697 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.56985699 - `BetaManagedAgentsWriteToolConfig object`57005701 Configuration for the write tool.57025703 - `type: "write"`57045705 - `enabled: boolean`57065707 - `name: "write"`57085709 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`57105711 Permission policy for tool execution.57125713 - `BetaManagedAgentsAlwaysAllowPolicy object`57145715 Tool calls are automatically approved without user confirmation.57165717 - `BetaManagedAgentsAlwaysAskPolicy object`57185719 Tool calls require user confirmation before execution.57205721 - `BetaManagedAgentsAutoPolicy object`57225723 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.57245725 - `BetaManagedAgentsGlobToolConfig object`57265727 Configuration for the glob tool.57285729 - `type: "glob"`57305731 - `enabled: boolean`57325733 - `name: "glob"`57345735 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`57365737 Permission policy for tool execution.57385739 - `BetaManagedAgentsAlwaysAllowPolicy object`57405741 Tool calls are automatically approved without user confirmation.57425743 - `BetaManagedAgentsAlwaysAskPolicy object`57445745 Tool calls require user confirmation before execution.57465747 - `BetaManagedAgentsAutoPolicy object`57485749 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.57505751 - `BetaManagedAgentsGrepToolConfig object`57525753 Configuration for the grep tool.57545755 - `type: "grep"`57565757 - `enabled: boolean`57585759 - `name: "grep"`57605761 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`57625763 Permission policy for tool execution.57645765 - `BetaManagedAgentsAlwaysAllowPolicy object`57665767 Tool calls are automatically approved without user confirmation.57685769 - `BetaManagedAgentsAlwaysAskPolicy object`57705771 Tool calls require user confirmation before execution.57725773 - `BetaManagedAgentsAutoPolicy object`57745775 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.57765777 - `BetaManagedAgentsWebFetchToolConfig object`57785779 Configuration for the web_fetch tool.57805781 - `type: "web_fetch"`57825783 - `enabled: boolean`57845785 - `name: "web_fetch"`57865787 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`57885789 Permission policy for tool execution.57905791 - `BetaManagedAgentsAlwaysAllowPolicy object`57925793 Tool calls are automatically approved without user confirmation.57945795 - `BetaManagedAgentsAlwaysAskPolicy object`57965797 Tool calls require user confirmation before execution.57985799 - `BetaManagedAgentsAutoPolicy object`58005801 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.58025803 - `allowed_domains: optional array of string`58045805 - `blocked_domains: optional array of string`58065807 - `max_content_tokens: optional number or null`58085809 format: int3258105811 - `BetaManagedAgentsWebSearchToolConfig object`58125813 Configuration for the web_search tool.58145815 - `type: "web_search"`58165817 - `enabled: boolean`58185819 - `name: "web_search"`58205821 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`58225823 Permission policy for tool execution.58245825 - `BetaManagedAgentsAlwaysAllowPolicy object`58265827 Tool calls are automatically approved without user confirmation.58285829 - `BetaManagedAgentsAlwaysAskPolicy object`58305831 Tool calls require user confirmation before execution.58325833 - `BetaManagedAgentsAutoPolicy object`58345835 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.58365837 - `allowed_domains: optional array of string`58385839 - `blocked_domains: optional array of string`58405841 - `user_location: optional BetaManagedAgentsUserLocation or null`58425843 Approximate user location for search result localization.58445845 - `type: "approximate"`58465847 Location precision. Only "approximate" is supported.58485849 - `city: optional string or null`58505851 City name.58525853 minLength: 1, maxLength: 25558545855 - `country: optional string or null`58565857 Two-letter ISO 3166-1 country code, uppercase.58585859 - `region: optional string or null`58605861 Region or state name.58625863 minLength: 1, maxLength: 25558645865 - `timezone: optional string or null`58665867 IANA timezone identifier, e.g. "America/Los_Angeles".58685869 minLength: 1, maxLength: 25558705871 - `default_config: BetaManagedAgentsAgentToolsetDefaultConfig`58725873 Resolved default configuration for agent tools.58745875 - `enabled: boolean`58765877 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`58785879 Permission policy for tool execution.58805881 - `BetaManagedAgentsAlwaysAllowPolicy object`58825883 Tool calls are automatically approved without user confirmation.58845885 - `BetaManagedAgentsAlwaysAskPolicy object`58865887 Tool calls require user confirmation before execution.58885889 - `BetaManagedAgentsAutoPolicy object`58905891 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.58925893 - `BetaManagedAgentsMCPToolset object`58945895 - `type: "mcp_toolset"`58965897 - `configs: array of BetaManagedAgentsMCPToolConfig`58985899 - `enabled: boolean`59005901 - `name: string`59025903 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`59045905 Permission policy for tool execution.59065907 - `BetaManagedAgentsAlwaysAllowPolicy object`59085909 Tool calls are automatically approved without user confirmation.59105911 - `BetaManagedAgentsAlwaysAskPolicy object`59125913 Tool calls require user confirmation before execution.59145915 - `BetaManagedAgentsAutoPolicy object`59165917 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.59185919 - `default_config: BetaManagedAgentsMCPToolsetDefaultConfig`59205921 Resolved default configuration for all tools from an MCP server.59225923 - `enabled: boolean`59245925 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`59265927 Permission policy for tool execution.59285929 - `BetaManagedAgentsAlwaysAllowPolicy object`59305931 Tool calls are automatically approved without user confirmation.59325933 - `BetaManagedAgentsAlwaysAskPolicy object`59345935 Tool calls require user confirmation before execution.59365937 - `BetaManagedAgentsAutoPolicy object`59385939 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.59405941 - `mcp_server_name: string`59425943 - `BetaManagedAgentsCustomTool object`59445945 A custom tool as returned in API responses.59465947 - `type: "custom"`59485949 - `description: string`59505951 - `input_schema: BetaManagedAgentsCustomToolInputSchema`59525953 JSON Schema for custom tool input parameters.59545955 - `type: "object"`59565957 - `properties: optional map[unknown] or null`59585959 - `required: optional array of string or null`59605961 - `name: string`59625963 - `updated_at: string`59645965 A timestamp in RFC 3339 format59665967 format: date-time59685969 - `version: number`59705971 The agent's current version. Starts at 1 and increments when the agent is modified.59725973 format: int3259745975### Beta Managed Agents Agent Reference59765977- `BetaManagedAgentsAgentReference object`59785979 A resolved agent reference with a concrete version.59805981 - `type: "agent"`59825983 - `id: string`59845985 - `version: number`59865987 format: int3259885989### Beta Managed Agents Agent Tool Config59905991- `BetaManagedAgentsAgentToolConfig = BetaManagedAgentsBashToolConfig or BetaManagedAgentsEditToolConfig or BetaManagedAgentsReadToolConfig or 5 more`59925993 Configuration for a specific agent tool.59945995 - `BetaManagedAgentsBashToolConfig object`59965997 Configuration for the bash tool.59985999 - `type: "bash"`60006001 - `enabled: boolean`60026003 - `name: "bash"`60046005 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`60066007 Permission policy for tool execution.60086009 - `BetaManagedAgentsAlwaysAllowPolicy object`60106011 Tool calls are automatically approved without user confirmation.60126013 - `type: "always_allow"`60146015 - `BetaManagedAgentsAlwaysAskPolicy object`60166017 Tool calls require user confirmation before execution.60186019 - `type: "always_ask"`60206021 - `BetaManagedAgentsAutoPolicy object`60226023 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.60246025 - `type: "auto"`60266027 - `BetaManagedAgentsEditToolConfig object`60286029 Configuration for the edit tool.60306031 - `type: "edit"`60326033 - `enabled: boolean`60346035 - `name: "edit"`60366037 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`60386039 Permission policy for tool execution.60406041 - `BetaManagedAgentsAlwaysAllowPolicy object`60426043 Tool calls are automatically approved without user confirmation.60446045 - `BetaManagedAgentsAlwaysAskPolicy object`60466047 Tool calls require user confirmation before execution.60486049 - `BetaManagedAgentsAutoPolicy object`60506051 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.60526053 - `BetaManagedAgentsReadToolConfig object`60546055 Configuration for the read tool.60566057 - `type: "read"`60586059 - `enabled: boolean`60606061 - `name: "read"`60626063 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`60646065 Permission policy for tool execution.60666067 - `BetaManagedAgentsAlwaysAllowPolicy object`60686069 Tool calls are automatically approved without user confirmation.60706071 - `BetaManagedAgentsAlwaysAskPolicy object`60726073 Tool calls require user confirmation before execution.60746075 - `BetaManagedAgentsAutoPolicy object`60766077 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.60786079 - `BetaManagedAgentsWriteToolConfig object`60806081 Configuration for the write tool.60826083 - `type: "write"`60846085 - `enabled: boolean`60866087 - `name: "write"`60886089 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`60906091 Permission policy for tool execution.60926093 - `BetaManagedAgentsAlwaysAllowPolicy object`60946095 Tool calls are automatically approved without user confirmation.60966097 - `BetaManagedAgentsAlwaysAskPolicy object`60986099 Tool calls require user confirmation before execution.61006101 - `BetaManagedAgentsAutoPolicy object`61026103 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.61046105 - `BetaManagedAgentsGlobToolConfig object`61066107 Configuration for the glob tool.61086109 - `type: "glob"`61106111 - `enabled: boolean`61126113 - `name: "glob"`61146115 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`61166117 Permission policy for tool execution.61186119 - `BetaManagedAgentsAlwaysAllowPolicy object`61206121 Tool calls are automatically approved without user confirmation.61226123 - `BetaManagedAgentsAlwaysAskPolicy object`61246125 Tool calls require user confirmation before execution.61266127 - `BetaManagedAgentsAutoPolicy object`61286129 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.61306131 - `BetaManagedAgentsGrepToolConfig object`61326133 Configuration for the grep tool.61346135 - `type: "grep"`61366137 - `enabled: boolean`61386139 - `name: "grep"`61406141 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`61426143 Permission policy for tool execution.61446145 - `BetaManagedAgentsAlwaysAllowPolicy object`61466147 Tool calls are automatically approved without user confirmation.61486149 - `BetaManagedAgentsAlwaysAskPolicy object`61506151 Tool calls require user confirmation before execution.61526153 - `BetaManagedAgentsAutoPolicy object`61546155 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.61566157 - `BetaManagedAgentsWebFetchToolConfig object`61586159 Configuration for the web_fetch tool.61606161 - `type: "web_fetch"`61626163 - `enabled: boolean`61646165 - `name: "web_fetch"`61666167 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`61686169 Permission policy for tool execution.61706171 - `BetaManagedAgentsAlwaysAllowPolicy object`61726173 Tool calls are automatically approved without user confirmation.61746175 - `BetaManagedAgentsAlwaysAskPolicy object`61766177 Tool calls require user confirmation before execution.61786179 - `BetaManagedAgentsAutoPolicy object`61806181 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.61826183 - `allowed_domains: optional array of string`61846185 - `blocked_domains: optional array of string`61866187 - `max_content_tokens: optional number or null`61886189 format: int3261906191 - `BetaManagedAgentsWebSearchToolConfig object`61926193 Configuration for the web_search tool.61946195 - `type: "web_search"`61966197 - `enabled: boolean`61986199 - `name: "web_search"`62006201 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`62026203 Permission policy for tool execution.62046205 - `BetaManagedAgentsAlwaysAllowPolicy object`62066207 Tool calls are automatically approved without user confirmation.62086209 - `BetaManagedAgentsAlwaysAskPolicy object`62106211 Tool calls require user confirmation before execution.62126213 - `BetaManagedAgentsAutoPolicy object`62146215 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.62166217 - `allowed_domains: optional array of string`62186219 - `blocked_domains: optional array of string`62206221 - `user_location: optional BetaManagedAgentsUserLocation or null`62226223 Approximate user location for search result localization.62246225 - `type: "approximate"`62266227 Location precision. Only "approximate" is supported.62286229 - `city: optional string or null`62306231 City name.62326233 minLength: 1, maxLength: 25562346235 - `country: optional string or null`62366237 Two-letter ISO 3166-1 country code, uppercase.62386239 - `region: optional string or null`62406241 Region or state name.62426243 minLength: 1, maxLength: 25562446245 - `timezone: optional string or null`62466247 IANA timezone identifier, e.g. "America/Los_Angeles".62486249 minLength: 1, maxLength: 25562506251### Beta Managed Agents Agent Tool Config Params62526253- `BetaManagedAgentsAgentToolConfigParams = BetaManagedAgentsBashToolConfigParams or BetaManagedAgentsEditToolConfigParams or BetaManagedAgentsReadToolConfigParams or 5 more`62546255 Configuration override for a specific tool within a toolset.62566257 - `BetaManagedAgentsBashToolConfigParams object`62586259 Configuration override for the bash tool.62606261 - `type: optional "bash"`62626263 - `name: "bash"`62646265 Must be "bash".62666267 - `enabled: optional boolean or null`62686269 Whether this tool is enabled and available to Claude. Overrides the default_config setting.62706271 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`62726273 Permission policy for tool execution.62746275 - `BetaManagedAgentsAlwaysAllowPolicy object`62766277 Tool calls are automatically approved without user confirmation.62786279 - `type: "always_allow"`62806281 - `BetaManagedAgentsAlwaysAskPolicy object`62826283 Tool calls require user confirmation before execution.62846285 - `type: "always_ask"`62866287 - `BetaManagedAgentsAutoPolicy object`62886289 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.62906291 - `type: "auto"`62926293 - `BetaManagedAgentsEditToolConfigParams object`62946295 Configuration override for the edit tool.62966297 - `type: optional "edit"`62986299 - `name: "edit"`63006301 Must be "edit".63026303 - `enabled: optional boolean or null`63046305 Whether this tool is enabled and available to Claude. Overrides the default_config setting.63066307 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`63086309 Permission policy for tool execution.63106311 - `BetaManagedAgentsAlwaysAllowPolicy object`63126313 Tool calls are automatically approved without user confirmation.63146315 - `BetaManagedAgentsAlwaysAskPolicy object`63166317 Tool calls require user confirmation before execution.63186319 - `BetaManagedAgentsAutoPolicy object`63206321 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.63226323 - `BetaManagedAgentsReadToolConfigParams object`63246325 Configuration override for the read tool.63266327 - `type: optional "read"`63286329 - `name: "read"`63306331 Must be "read".63326333 - `enabled: optional boolean or null`63346335 Whether this tool is enabled and available to Claude. Overrides the default_config setting.63366337 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`63386339 Permission policy for tool execution.63406341 - `BetaManagedAgentsAlwaysAllowPolicy object`63426343 Tool calls are automatically approved without user confirmation.63446345 - `BetaManagedAgentsAlwaysAskPolicy object`63466347 Tool calls require user confirmation before execution.63486349 - `BetaManagedAgentsAutoPolicy object`63506351 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.63526353 - `BetaManagedAgentsWriteToolConfigParams object`63546355 Configuration override for the write tool.63566357 - `type: optional "write"`63586359 - `name: "write"`63606361 Must be "write".63626363 - `enabled: optional boolean or null`63646365 Whether this tool is enabled and available to Claude. Overrides the default_config setting.63666367 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`63686369 Permission policy for tool execution.63706371 - `BetaManagedAgentsAlwaysAllowPolicy object`63726373 Tool calls are automatically approved without user confirmation.63746375 - `BetaManagedAgentsAlwaysAskPolicy object`63766377 Tool calls require user confirmation before execution.63786379 - `BetaManagedAgentsAutoPolicy object`63806381 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.63826383 - `BetaManagedAgentsGlobToolConfigParams object`63846385 Configuration override for the glob tool.63866387 - `type: optional "glob"`63886389 - `name: "glob"`63906391 Must be "glob".63926393 - `enabled: optional boolean or null`63946395 Whether this tool is enabled and available to Claude. Overrides the default_config setting.63966397 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`63986399 Permission policy for tool execution.64006401 - `BetaManagedAgentsAlwaysAllowPolicy object`64026403 Tool calls are automatically approved without user confirmation.64046405 - `BetaManagedAgentsAlwaysAskPolicy object`64066407 Tool calls require user confirmation before execution.64086409 - `BetaManagedAgentsAutoPolicy object`64106411 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.64126413 - `BetaManagedAgentsGrepToolConfigParams object`64146415 Configuration override for the grep tool.64166417 - `type: optional "grep"`64186419 - `name: "grep"`64206421 Must be "grep".64226423 - `enabled: optional boolean or null`64246425 Whether this tool is enabled and available to Claude. Overrides the default_config setting.64266427 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`64286429 Permission policy for tool execution.64306431 - `BetaManagedAgentsAlwaysAllowPolicy object`64326433 Tool calls are automatically approved without user confirmation.64346435 - `BetaManagedAgentsAlwaysAskPolicy object`64366437 Tool calls require user confirmation before execution.64386439 - `BetaManagedAgentsAutoPolicy object`64406441 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.64426443 - `BetaManagedAgentsWebFetchToolConfigParams object`64446445 Configuration override for the web_fetch tool.64466447 - `type: optional "web_fetch"`64486449 - `name: "web_fetch"`64506451 Must be "web_fetch".64526453 - `allowed_domains: optional array of string`64546455 Only fetch URLs whose host is one of these domains or a subdomain of one. Each entry is a plain hostname like "docs.example.com" (no scheme, port, or path). At most 64 entries; an empty list is rejected (omit the field instead). Cannot be combined with blocked_domains.64566457 - `blocked_domains: optional array of string`64586459 Never fetch URLs whose host is one of these domains or a subdomain of one. Each entry is a plain hostname like "ads.example.com" (no scheme, port, or path). At most 64 entries; an empty list is rejected (omit the field instead). Cannot be combined with allowed_domains.64606461 - `enabled: optional boolean or null`64626463 Whether this tool is enabled and available to Claude. Overrides the default_config setting.64646465 - `max_content_tokens: optional number or null`64666467 Maximum number of tokens of fetched text content to include in context per call. Does not apply to binary content such as PDFs.64686469 format: int3264706471 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`64726473 Permission policy for tool execution.64746475 - `BetaManagedAgentsAlwaysAllowPolicy object`64766477 Tool calls are automatically approved without user confirmation.64786479 - `BetaManagedAgentsAlwaysAskPolicy object`64806481 Tool calls require user confirmation before execution.64826483 - `BetaManagedAgentsAutoPolicy object`64846485 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.64866487 - `BetaManagedAgentsWebSearchToolConfigParams object`64886489 Configuration override for the web_search tool.64906491 - `type: optional "web_search"`64926493 - `name: "web_search"`64946495 Must be "web_search".64966497 - `allowed_domains: optional array of string`64986499 Only return search results whose host is one of these domains or a subdomain of one. Each entry is a plain hostname like "docs.example.com" (no scheme or port; an optional path suffix is accepted). At most 64 entries; an empty list is rejected (omit the field instead). Cannot be combined with blocked_domains.65006501 - `blocked_domains: optional array of string`65026503 Never return search results whose host is one of these domains or a subdomain of one. Each entry is a plain hostname like "ads.example.com" (no scheme or port; an optional path suffix is accepted). At most 64 entries; an empty list is rejected (omit the field instead). Cannot be combined with allowed_domains.65046505 - `enabled: optional boolean or null`65066507 Whether this tool is enabled and available to Claude. Overrides the default_config setting.65086509 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`65106511 Permission policy for tool execution.65126513 - `BetaManagedAgentsAlwaysAllowPolicy object`65146515 Tool calls are automatically approved without user confirmation.65166517 - `BetaManagedAgentsAlwaysAskPolicy object`65186519 Tool calls require user confirmation before execution.65206521 - `BetaManagedAgentsAutoPolicy object`65226523 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.65246525 - `user_location: optional BetaManagedAgentsUserLocation or null`65266527 Approximate user location for search result localization.65286529 - `type: "approximate"`65306531 Location precision. Only "approximate" is supported.65326533 - `city: optional string or null`65346535 City name.65366537 minLength: 1, maxLength: 25565386539 - `country: optional string or null`65406541 Two-letter ISO 3166-1 country code, uppercase.65426543 - `region: optional string or null`65446545 Region or state name.65466547 minLength: 1, maxLength: 25565486549 - `timezone: optional string or null`65506551 IANA timezone identifier, e.g. "America/Los_Angeles".65526553 minLength: 1, maxLength: 25565546555### Beta Managed Agents Agent Toolset Default Config65566557- `BetaManagedAgentsAgentToolsetDefaultConfig object`65586559 Resolved default configuration for agent tools.65606561 - `enabled: boolean`65626563 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`65646565 Permission policy for tool execution.65666567 - `BetaManagedAgentsAlwaysAllowPolicy object`65686569 Tool calls are automatically approved without user confirmation.65706571 - `type: "always_allow"`65726573 - `BetaManagedAgentsAlwaysAskPolicy object`65746575 Tool calls require user confirmation before execution.65766577 - `type: "always_ask"`65786579 - `BetaManagedAgentsAutoPolicy object`65806581 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.65826583 - `type: "auto"`65846585### Beta Managed Agents Agent Toolset Default Config Params65866587- `BetaManagedAgentsAgentToolsetDefaultConfigParams object`65886589 Default configuration for all tools in a toolset.65906591 - `enabled: optional boolean or null`65926593 Whether tools are enabled and available to Claude by default. Defaults to true if not specified.65946595 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`65966597 Permission policy for tool execution.65986599 - `BetaManagedAgentsAlwaysAllowPolicy object`66006601 Tool calls are automatically approved without user confirmation.66026603 - `type: "always_allow"`66046605 - `BetaManagedAgentsAlwaysAskPolicy object`66066607 Tool calls require user confirmation before execution.66086609 - `type: "always_ask"`66106611 - `BetaManagedAgentsAutoPolicy object`66126613 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.66146615 - `type: "auto"`66166617### Beta Managed Agents Agent Toolset2026040166186619- `BetaManagedAgentsAgentToolset20260401 object`66206621 - `type: "agent_toolset_20260401"`66226623 - `configs: array of BetaManagedAgentsAgentToolConfig`66246625 - `BetaManagedAgentsBashToolConfig object`66266627 Configuration for the bash tool.66286629 - `type: "bash"`66306631 - `enabled: boolean`66326633 - `name: "bash"`66346635 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`66366637 Permission policy for tool execution.66386639 - `BetaManagedAgentsAlwaysAllowPolicy object`66406641 Tool calls are automatically approved without user confirmation.66426643 - `type: "always_allow"`66446645 - `BetaManagedAgentsAlwaysAskPolicy object`66466647 Tool calls require user confirmation before execution.66486649 - `type: "always_ask"`66506651 - `BetaManagedAgentsAutoPolicy object`66526653 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.66546655 - `type: "auto"`66566657 - `BetaManagedAgentsEditToolConfig object`66586659 Configuration for the edit tool.66606661 - `type: "edit"`66626663 - `enabled: boolean`66646665 - `name: "edit"`66666667 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`66686669 Permission policy for tool execution.66706671 - `BetaManagedAgentsAlwaysAllowPolicy object`66726673 Tool calls are automatically approved without user confirmation.66746675 - `BetaManagedAgentsAlwaysAskPolicy object`66766677 Tool calls require user confirmation before execution.66786679 - `BetaManagedAgentsAutoPolicy object`66806681 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.66826683 - `BetaManagedAgentsReadToolConfig object`66846685 Configuration for the read tool.66866687 - `type: "read"`66886689 - `enabled: boolean`66906691 - `name: "read"`66926693 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`66946695 Permission policy for tool execution.66966697 - `BetaManagedAgentsAlwaysAllowPolicy object`66986699 Tool calls are automatically approved without user confirmation.67006701 - `BetaManagedAgentsAlwaysAskPolicy object`67026703 Tool calls require user confirmation before execution.67046705 - `BetaManagedAgentsAutoPolicy object`67066707 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.67086709 - `BetaManagedAgentsWriteToolConfig object`67106711 Configuration for the write tool.67126713 - `type: "write"`67146715 - `enabled: boolean`67166717 - `name: "write"`67186719 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`67206721 Permission policy for tool execution.67226723 - `BetaManagedAgentsAlwaysAllowPolicy object`67246725 Tool calls are automatically approved without user confirmation.67266727 - `BetaManagedAgentsAlwaysAskPolicy object`67286729 Tool calls require user confirmation before execution.67306731 - `BetaManagedAgentsAutoPolicy object`67326733 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.67346735 - `BetaManagedAgentsGlobToolConfig object`67366737 Configuration for the glob tool.67386739 - `type: "glob"`67406741 - `enabled: boolean`67426743 - `name: "glob"`67446745 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`67466747 Permission policy for tool execution.67486749 - `BetaManagedAgentsAlwaysAllowPolicy object`67506751 Tool calls are automatically approved without user confirmation.67526753 - `BetaManagedAgentsAlwaysAskPolicy object`67546755 Tool calls require user confirmation before execution.67566757 - `BetaManagedAgentsAutoPolicy object`67586759 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.67606761 - `BetaManagedAgentsGrepToolConfig object`67626763 Configuration for the grep tool.67646765 - `type: "grep"`67666767 - `enabled: boolean`67686769 - `name: "grep"`67706771 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`67726773 Permission policy for tool execution.67746775 - `BetaManagedAgentsAlwaysAllowPolicy object`67766777 Tool calls are automatically approved without user confirmation.67786779 - `BetaManagedAgentsAlwaysAskPolicy object`67806781 Tool calls require user confirmation before execution.67826783 - `BetaManagedAgentsAutoPolicy object`67846785 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.67866787 - `BetaManagedAgentsWebFetchToolConfig object`67886789 Configuration for the web_fetch tool.67906791 - `type: "web_fetch"`67926793 - `enabled: boolean`67946795 - `name: "web_fetch"`67966797 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`67986799 Permission policy for tool execution.68006801 - `BetaManagedAgentsAlwaysAllowPolicy object`68026803 Tool calls are automatically approved without user confirmation.68046805 - `BetaManagedAgentsAlwaysAskPolicy object`68066807 Tool calls require user confirmation before execution.68086809 - `BetaManagedAgentsAutoPolicy object`68106811 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.68126813 - `allowed_domains: optional array of string`68146815 - `blocked_domains: optional array of string`68166817 - `max_content_tokens: optional number or null`68186819 format: int3268206821 - `BetaManagedAgentsWebSearchToolConfig object`68226823 Configuration for the web_search tool.68246825 - `type: "web_search"`68266827 - `enabled: boolean`68286829 - `name: "web_search"`68306831 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`68326833 Permission policy for tool execution.68346835 - `BetaManagedAgentsAlwaysAllowPolicy object`68366837 Tool calls are automatically approved without user confirmation.68386839 - `BetaManagedAgentsAlwaysAskPolicy object`68406841 Tool calls require user confirmation before execution.68426843 - `BetaManagedAgentsAutoPolicy object`68446845 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.68466847 - `allowed_domains: optional array of string`68486849 - `blocked_domains: optional array of string`68506851 - `user_location: optional BetaManagedAgentsUserLocation or null`68526853 Approximate user location for search result localization.68546855 - `type: "approximate"`68566857 Location precision. Only "approximate" is supported.68586859 - `city: optional string or null`68606861 City name.68626863 minLength: 1, maxLength: 25568646865 - `country: optional string or null`68666867 Two-letter ISO 3166-1 country code, uppercase.68686869 - `region: optional string or null`68706871 Region or state name.68726873 minLength: 1, maxLength: 25568746875 - `timezone: optional string or null`68766877 IANA timezone identifier, e.g. "America/Los_Angeles".68786879 minLength: 1, maxLength: 25568806881 - `default_config: BetaManagedAgentsAgentToolsetDefaultConfig`68826883 Resolved default configuration for agent tools.68846885 - `enabled: boolean`68866887 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`68886889 Permission policy for tool execution.68906891 - `BetaManagedAgentsAlwaysAllowPolicy object`68926893 Tool calls are automatically approved without user confirmation.68946895 - `BetaManagedAgentsAlwaysAskPolicy object`68966897 Tool calls require user confirmation before execution.68986899 - `BetaManagedAgentsAutoPolicy object`69006901 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.69026903### Beta Managed Agents Agent Toolset20260401 Bash Input69046905- `BetaManagedAgentsAgentToolset20260401BashInput object`69066907 Input payload for the `bash` tool of the6908 `agent_toolset_20260401` toolset. All fields are optional;6909 a normal invocation supplies `command`, while `restart=true`6910 (with no `command`) reboots the runner-side bash session.69116912 - `command: optional string`69136914 Shell command to execute. Omit only when `restart` is true.69156916 - `restart: optional boolean`69176918 When true, restart the persistent bash session instead of6919 running a command. Subsequent calls without `restart` will6920 run against the fresh session.69216922 - `timeout_ms: optional number`69236924 Per-call timeout in milliseconds. Defaults to the6925 runner-wide tool timeout when omitted or zero.69266927 minimum: 069286929### Beta Managed Agents Agent Toolset20260401 Edit Input69306931- `BetaManagedAgentsAgentToolset20260401EditInput object`69326933 Input payload for the `edit` tool. Performs a string6934 replacement in the named file; by default `old_string` must6935 occur exactly once.69366937 - `file_path: string`69386939 Path of the file to edit.69406941 - `new_string: string`69426943 Replacement text.69446945 - `old_string: string`69466947 Substring to find and replace.69486949 - `replace_all: optional boolean`69506951 When true, replace every occurrence of `old_string`6952 instead of requiring a unique match.69536954### Beta Managed Agents Agent Toolset20260401 Glob Input69556956- `BetaManagedAgentsAgentToolset20260401GlobInput object`69576958 Input payload for the `glob` tool. Returns paths matching a6959 doublestar glob pattern, newest first.69606961 - `pattern: string`69626963 Doublestar glob pattern (e.g. `**/*.go`). Absolute patterns6964 are only permitted when the runner is configured to allow6965 them.69666967 - `path: optional string`69686969 Optional directory root to search under. Defaults to the6970 runner's working directory.69716972### Beta Managed Agents Agent Toolset20260401 Grep Input69736974- `BetaManagedAgentsAgentToolset20260401GrepInput object`69756976 Input payload for the `grep` tool. Searches file contents for6977 a regular expression, returning matching lines.69786979 - `pattern: string`69806981 Regular expression to search for.69826983 - `path: optional string`69846985 Optional directory root to search under. Defaults to the6986 runner's working directory.69876988### Beta Managed Agents Agent Toolset20260401 Params69896990- `BetaManagedAgentsAgentToolset20260401Params object`69916992 Configuration for built-in agent tools. Use this to enable or disable groups of tools available to the agent.69936994 - `type: "agent_toolset_20260401"`69956996 - `configs: optional array of BetaManagedAgentsAgentToolConfigParams`69976998 Per-tool configuration overrides.69997000 - `BetaManagedAgentsBashToolConfigParams object`70017002 Configuration override for the bash tool.70037004 - `type: optional "bash"`70057006 - `name: "bash"`70077008 Must be "bash".70097010 - `enabled: optional boolean or null`70117012 Whether this tool is enabled and available to Claude. Overrides the default_config setting.70137014 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`70157016 Permission policy for tool execution.70177018 - `BetaManagedAgentsAlwaysAllowPolicy object`70197020 Tool calls are automatically approved without user confirmation.70217022 - `type: "always_allow"`70237024 - `BetaManagedAgentsAlwaysAskPolicy object`70257026 Tool calls require user confirmation before execution.70277028 - `type: "always_ask"`70297030 - `BetaManagedAgentsAutoPolicy object`70317032 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.70337034 - `type: "auto"`70357036 - `BetaManagedAgentsEditToolConfigParams object`70377038 Configuration override for the edit tool.70397040 - `type: optional "edit"`70417042 - `name: "edit"`70437044 Must be "edit".70457046 - `enabled: optional boolean or null`70477048 Whether this tool is enabled and available to Claude. Overrides the default_config setting.70497050 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`70517052 Permission policy for tool execution.70537054 - `BetaManagedAgentsAlwaysAllowPolicy object`70557056 Tool calls are automatically approved without user confirmation.70577058 - `BetaManagedAgentsAlwaysAskPolicy object`70597060 Tool calls require user confirmation before execution.70617062 - `BetaManagedAgentsAutoPolicy object`70637064 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.70657066 - `BetaManagedAgentsReadToolConfigParams object`70677068 Configuration override for the read tool.70697070 - `type: optional "read"`70717072 - `name: "read"`70737074 Must be "read".70757076 - `enabled: optional boolean or null`70777078 Whether this tool is enabled and available to Claude. Overrides the default_config setting.70797080 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`70817082 Permission policy for tool execution.70837084 - `BetaManagedAgentsAlwaysAllowPolicy object`70857086 Tool calls are automatically approved without user confirmation.70877088 - `BetaManagedAgentsAlwaysAskPolicy object`70897090 Tool calls require user confirmation before execution.70917092 - `BetaManagedAgentsAutoPolicy object`70937094 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.70957096 - `BetaManagedAgentsWriteToolConfigParams object`70977098 Configuration override for the write tool.70997100 - `type: optional "write"`71017102 - `name: "write"`71037104 Must be "write".71057106 - `enabled: optional boolean or null`71077108 Whether this tool is enabled and available to Claude. Overrides the default_config setting.71097110 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`71117112 Permission policy for tool execution.71137114 - `BetaManagedAgentsAlwaysAllowPolicy object`71157116 Tool calls are automatically approved without user confirmation.71177118 - `BetaManagedAgentsAlwaysAskPolicy object`71197120 Tool calls require user confirmation before execution.71217122 - `BetaManagedAgentsAutoPolicy object`71237124 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.71257126 - `BetaManagedAgentsGlobToolConfigParams object`71277128 Configuration override for the glob tool.71297130 - `type: optional "glob"`71317132 - `name: "glob"`71337134 Must be "glob".71357136 - `enabled: optional boolean or null`71377138 Whether this tool is enabled and available to Claude. Overrides the default_config setting.71397140 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`71417142 Permission policy for tool execution.71437144 - `BetaManagedAgentsAlwaysAllowPolicy object`71457146 Tool calls are automatically approved without user confirmation.71477148 - `BetaManagedAgentsAlwaysAskPolicy object`71497150 Tool calls require user confirmation before execution.71517152 - `BetaManagedAgentsAutoPolicy object`71537154 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.71557156 - `BetaManagedAgentsGrepToolConfigParams object`71577158 Configuration override for the grep tool.71597160 - `type: optional "grep"`71617162 - `name: "grep"`71637164 Must be "grep".71657166 - `enabled: optional boolean or null`71677168 Whether this tool is enabled and available to Claude. Overrides the default_config setting.71697170 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`71717172 Permission policy for tool execution.71737174 - `BetaManagedAgentsAlwaysAllowPolicy object`71757176 Tool calls are automatically approved without user confirmation.71777178 - `BetaManagedAgentsAlwaysAskPolicy object`71797180 Tool calls require user confirmation before execution.71817182 - `BetaManagedAgentsAutoPolicy object`71837184 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.71857186 - `BetaManagedAgentsWebFetchToolConfigParams object`71877188 Configuration override for the web_fetch tool.71897190 - `type: optional "web_fetch"`71917192 - `name: "web_fetch"`71937194 Must be "web_fetch".71957196 - `allowed_domains: optional array of string`71977198 Only fetch URLs whose host is one of these domains or a subdomain of one. Each entry is a plain hostname like "docs.example.com" (no scheme, port, or path). At most 64 entries; an empty list is rejected (omit the field instead). Cannot be combined with blocked_domains.71997200 - `blocked_domains: optional array of string`72017202 Never fetch URLs whose host is one of these domains or a subdomain of one. Each entry is a plain hostname like "ads.example.com" (no scheme, port, or path). At most 64 entries; an empty list is rejected (omit the field instead). Cannot be combined with allowed_domains.72037204 - `enabled: optional boolean or null`72057206 Whether this tool is enabled and available to Claude. Overrides the default_config setting.72077208 - `max_content_tokens: optional number or null`72097210 Maximum number of tokens of fetched text content to include in context per call. Does not apply to binary content such as PDFs.72117212 format: int3272137214 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`72157216 Permission policy for tool execution.72177218 - `BetaManagedAgentsAlwaysAllowPolicy object`72197220 Tool calls are automatically approved without user confirmation.72217222 - `BetaManagedAgentsAlwaysAskPolicy object`72237224 Tool calls require user confirmation before execution.72257226 - `BetaManagedAgentsAutoPolicy object`72277228 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.72297230 - `BetaManagedAgentsWebSearchToolConfigParams object`72317232 Configuration override for the web_search tool.72337234 - `type: optional "web_search"`72357236 - `name: "web_search"`72377238 Must be "web_search".72397240 - `allowed_domains: optional array of string`72417242 Only return search results whose host is one of these domains or a subdomain of one. Each entry is a plain hostname like "docs.example.com" (no scheme or port; an optional path suffix is accepted). At most 64 entries; an empty list is rejected (omit the field instead). Cannot be combined with blocked_domains.72437244 - `blocked_domains: optional array of string`72457246 Never return search results whose host is one of these domains or a subdomain of one. Each entry is a plain hostname like "ads.example.com" (no scheme or port; an optional path suffix is accepted). At most 64 entries; an empty list is rejected (omit the field instead). Cannot be combined with allowed_domains.72477248 - `enabled: optional boolean or null`72497250 Whether this tool is enabled and available to Claude. Overrides the default_config setting.72517252 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`72537254 Permission policy for tool execution.72557256 - `BetaManagedAgentsAlwaysAllowPolicy object`72577258 Tool calls are automatically approved without user confirmation.72597260 - `BetaManagedAgentsAlwaysAskPolicy object`72617262 Tool calls require user confirmation before execution.72637264 - `BetaManagedAgentsAutoPolicy object`72657266 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.72677268 - `user_location: optional BetaManagedAgentsUserLocation or null`72697270 Approximate user location for search result localization.72717272 - `type: "approximate"`72737274 Location precision. Only "approximate" is supported.72757276 - `city: optional string or null`72777278 City name.72797280 minLength: 1, maxLength: 25572817282 - `country: optional string or null`72837284 Two-letter ISO 3166-1 country code, uppercase.72857286 - `region: optional string or null`72877288 Region or state name.72897290 minLength: 1, maxLength: 25572917292 - `timezone: optional string or null`72937294 IANA timezone identifier, e.g. "America/Los_Angeles".72957296 minLength: 1, maxLength: 25572977298 - `default_config: optional BetaManagedAgentsAgentToolsetDefaultConfigParams or null`72997300 Default configuration for all tools in a toolset.73017302 - `enabled: optional boolean or null`73037304 Whether tools are enabled and available to Claude by default. Defaults to true if not specified.73057306 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`73077308 Permission policy for tool execution.73097310 - `BetaManagedAgentsAlwaysAllowPolicy object`73117312 Tool calls are automatically approved without user confirmation.73137314 - `BetaManagedAgentsAlwaysAskPolicy object`73157316 Tool calls require user confirmation before execution.73177318 - `BetaManagedAgentsAutoPolicy object`73197320 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.73217322### Beta Managed Agents Agent Toolset20260401 Read Input73237324- `BetaManagedAgentsAgentToolset20260401ReadInput object`73257326 Input payload for the `read` tool. Reads file contents7327 relative to the runner's working directory (or absolute when7328 the runner permits).73297330 - `file_path: string`73317332 Path of the file to read.73337334 - `view_range: optional array of number`73357336 Optional `[start_line, end_line]` 1-indexed inclusive7337 range. When omitted the entire file is returned.7338 `end_line` of 0 or negative means "to end of file".73397340 minItems: 2, maxItems: 273417342### Beta Managed Agents Agent Toolset20260401 Write Input73437344- `BetaManagedAgentsAgentToolset20260401WriteInput object`73457346 Input payload for the `write` tool. Writes (overwriting) the7347 entire file contents.73487349 - `content: string`73507351 Full file contents to write.73527353 - `file_path: string`73547355 Path of the file to write.73567357### Beta Managed Agents Always Allow Policy73587359- `BetaManagedAgentsAlwaysAllowPolicy object`73607361 Tool calls are automatically approved without user confirmation.73627363 - `type: "always_allow"`73647365### Beta Managed Agents Always Ask Policy73667367- `BetaManagedAgentsAlwaysAskPolicy object`73687369 Tool calls require user confirmation before execution.73707371 - `type: "always_ask"`73727373### Beta Managed Agents Anthropic Skill73747375- `BetaManagedAgentsAnthropicSkill object`73767377 A resolved Anthropic-managed skill.73787379 - `type: "anthropic"`73807381 - `skill_id: string`73827383 - `version: string`73847385### Beta Managed Agents Anthropic Skill Params73867387- `BetaManagedAgentsAnthropicSkillParams object`73887389 An Anthropic-managed skill.73907391 - `type: "anthropic"`73927393 - `skill_id: string`73947395 Identifier of the Anthropic skill (e.g., "xlsx").73967397 minLength: 1, maxLength: 6473987399 - `version: optional string or null`74007401 Version to pin. Defaults to latest if omitted.74027403 minLength: 1, maxLength: 6474047405### Beta Managed Agents Auto Policy74067407- `BetaManagedAgentsAutoPolicy object`74087409 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.74107411 - `type: "auto"`74127413### Beta Managed Agents Bash Tool Config74147415- `BetaManagedAgentsBashToolConfig object`74167417 Configuration for the bash tool.74187419 - `type: "bash"`74207421 - `enabled: boolean`74227423 - `name: "bash"`74247425 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`74267427 Permission policy for tool execution.74287429 - `BetaManagedAgentsAlwaysAllowPolicy object`74307431 Tool calls are automatically approved without user confirmation.74327433 - `type: "always_allow"`74347435 - `BetaManagedAgentsAlwaysAskPolicy object`74367437 Tool calls require user confirmation before execution.74387439 - `type: "always_ask"`74407441 - `BetaManagedAgentsAutoPolicy object`74427443 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.74447445 - `type: "auto"`74467447### Beta Managed Agents Bash Tool Config Params74487449- `BetaManagedAgentsBashToolConfigParams object`74507451 Configuration override for the bash tool.74527453 - `type: optional "bash"`74547455 - `name: "bash"`74567457 Must be "bash".74587459 - `enabled: optional boolean or null`74607461 Whether this tool is enabled and available to Claude. Overrides the default_config setting.74627463 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`74647465 Permission policy for tool execution.74667467 - `BetaManagedAgentsAlwaysAllowPolicy object`74687469 Tool calls are automatically approved without user confirmation.74707471 - `type: "always_allow"`74727473 - `BetaManagedAgentsAlwaysAskPolicy object`74747475 Tool calls require user confirmation before execution.74767477 - `type: "always_ask"`74787479 - `BetaManagedAgentsAutoPolicy object`74807481 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.74827483 - `type: "auto"`74847485### Beta Managed Agents Custom Skill74867487- `BetaManagedAgentsCustomSkill object`74887489 A resolved user-created custom skill.74907491 - `type: "custom"`74927493 - `skill_id: string`74947495 - `version: string`74967497### Beta Managed Agents Custom Skill Params74987499- `BetaManagedAgentsCustomSkillParams object`75007501 A user-created custom skill.75027503 - `type: "custom"`75047505 - `skill_id: string`75067507 Tagged ID of the custom skill (e.g., "skill_01XJ5...").75087509 minLength: 1, maxLength: 6475107511 - `version: optional string or null`75127513 Version to pin. Defaults to latest if omitted.75147515 minLength: 1, maxLength: 6475167517### Beta Managed Agents Custom Tool75187519- `BetaManagedAgentsCustomTool object`75207521 A custom tool as returned in API responses.75227523 - `type: "custom"`75247525 - `description: string`75267527 - `input_schema: BetaManagedAgentsCustomToolInputSchema`75287529 JSON Schema for custom tool input parameters.75307531 - `type: "object"`75327533 - `properties: optional map[unknown] or null`75347535 - `required: optional array of string or null`75367537 - `name: string`75387539### Beta Managed Agents Custom Tool Input Schema75407541- `BetaManagedAgentsCustomToolInputSchema object`75427543 JSON Schema for custom tool input parameters.75447545 - `type: "object"`75467547 - `properties: optional map[unknown] or null`75487549 - `required: optional array of string or null`75507551### Beta Managed Agents Custom Tool Params75527553- `BetaManagedAgentsCustomToolParams object`75547555 A custom tool that is executed by the API client rather than the agent. When the agent calls this tool, an `agent.custom_tool_use` event is emitted and the session goes idle, waiting for the client to provide the result via a `user.custom_tool_result` event.75567557 - `type: "custom"`75587559 - `description: string`75607561 Description of what the tool does, shown to the agent to help it decide when to use the tool.75627563 minLength: 175647565 - `input_schema: BetaManagedAgentsCustomToolInputSchema`75667567 JSON Schema for custom tool input parameters.75687569 - `type: "object"`75707571 - `properties: optional map[unknown] or null`75727573 - `required: optional array of string or null`75747575 - `name: string`75767577 Unique name for the tool. 1-128 characters; letters, digits, underscores, and hyphens.75787579 minLength: 1, maxLength: 12875807581### Beta Managed Agents Edit Tool Config75827583- `BetaManagedAgentsEditToolConfig object`75847585 Configuration for the edit tool.75867587 - `type: "edit"`75887589 - `enabled: boolean`75907591 - `name: "edit"`75927593 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`75947595 Permission policy for tool execution.75967597 - `BetaManagedAgentsAlwaysAllowPolicy object`75987599 Tool calls are automatically approved without user confirmation.76007601 - `type: "always_allow"`76027603 - `BetaManagedAgentsAlwaysAskPolicy object`76047605 Tool calls require user confirmation before execution.76067607 - `type: "always_ask"`76087609 - `BetaManagedAgentsAutoPolicy object`76107611 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.76127613 - `type: "auto"`76147615### Beta Managed Agents Edit Tool Config Params76167617- `BetaManagedAgentsEditToolConfigParams object`76187619 Configuration override for the edit tool.76207621 - `type: optional "edit"`76227623 - `name: "edit"`76247625 Must be "edit".76267627 - `enabled: optional boolean or null`76287629 Whether this tool is enabled and available to Claude. Overrides the default_config setting.76307631 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`76327633 Permission policy for tool execution.76347635 - `BetaManagedAgentsAlwaysAllowPolicy object`76367637 Tool calls are automatically approved without user confirmation.76387639 - `type: "always_allow"`76407641 - `BetaManagedAgentsAlwaysAskPolicy object`76427643 Tool calls require user confirmation before execution.76447645 - `type: "always_ask"`76467647 - `BetaManagedAgentsAutoPolicy object`76487649 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.76507651 - `type: "auto"`76527653### Beta Managed Agents Effort High76547655- `BetaManagedAgentsEffortHigh object`76567657 High effort. Favors reasoning depth.76587659 - `type: "high"`76607661### Beta Managed Agents Effort Low76627663- `BetaManagedAgentsEffortLow object`76647665 Low effort. Favors latency over reasoning depth.76667667 - `type: "low"`76687669### Beta Managed Agents Effort Max76707671- `BetaManagedAgentsEffortMax object`76727673 Maximum effort. Favors reasoning depth over latency.76747675 - `type: "max"`76767677### Beta Managed Agents Effort Medium76787679- `BetaManagedAgentsEffortMedium object`76807681 Medium effort. Balances latency and reasoning depth.76827683 - `type: "medium"`76847685### Beta Managed Agents Effort Xhigh76867687- `BetaManagedAgentsEffortXhigh object`76887689 Extra-high effort. Not all models accept this level.76907691 - `type: "xhigh"`76927693### Beta Managed Agents Glob Tool Config76947695- `BetaManagedAgentsGlobToolConfig object`76967697 Configuration for the glob tool.76987699 - `type: "glob"`77007701 - `enabled: boolean`77027703 - `name: "glob"`77047705 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`77067707 Permission policy for tool execution.77087709 - `BetaManagedAgentsAlwaysAllowPolicy object`77107711 Tool calls are automatically approved without user confirmation.77127713 - `type: "always_allow"`77147715 - `BetaManagedAgentsAlwaysAskPolicy object`77167717 Tool calls require user confirmation before execution.77187719 - `type: "always_ask"`77207721 - `BetaManagedAgentsAutoPolicy object`77227723 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.77247725 - `type: "auto"`77267727### Beta Managed Agents Glob Tool Config Params77287729- `BetaManagedAgentsGlobToolConfigParams object`77307731 Configuration override for the glob tool.77327733 - `type: optional "glob"`77347735 - `name: "glob"`77367737 Must be "glob".77387739 - `enabled: optional boolean or null`77407741 Whether this tool is enabled and available to Claude. Overrides the default_config setting.77427743 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`77447745 Permission policy for tool execution.77467747 - `BetaManagedAgentsAlwaysAllowPolicy object`77487749 Tool calls are automatically approved without user confirmation.77507751 - `type: "always_allow"`77527753 - `BetaManagedAgentsAlwaysAskPolicy object`77547755 Tool calls require user confirmation before execution.77567757 - `type: "always_ask"`77587759 - `BetaManagedAgentsAutoPolicy object`77607761 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.77627763 - `type: "auto"`77647765### Beta Managed Agents Grep Tool Config77667767- `BetaManagedAgentsGrepToolConfig object`77687769 Configuration for the grep tool.77707771 - `type: "grep"`77727773 - `enabled: boolean`77747775 - `name: "grep"`77767777 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`77787779 Permission policy for tool execution.77807781 - `BetaManagedAgentsAlwaysAllowPolicy object`77827783 Tool calls are automatically approved without user confirmation.77847785 - `type: "always_allow"`77867787 - `BetaManagedAgentsAlwaysAskPolicy object`77887789 Tool calls require user confirmation before execution.77907791 - `type: "always_ask"`77927793 - `BetaManagedAgentsAutoPolicy object`77947795 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.77967797 - `type: "auto"`77987799### Beta Managed Agents Grep Tool Config Params78007801- `BetaManagedAgentsGrepToolConfigParams object`78027803 Configuration override for the grep tool.78047805 - `type: optional "grep"`78067807 - `name: "grep"`78087809 Must be "grep".78107811 - `enabled: optional boolean or null`78127813 Whether this tool is enabled and available to Claude. Overrides the default_config setting.78147815 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`78167817 Permission policy for tool execution.78187819 - `BetaManagedAgentsAlwaysAllowPolicy object`78207821 Tool calls are automatically approved without user confirmation.78227823 - `type: "always_allow"`78247825 - `BetaManagedAgentsAlwaysAskPolicy object`78267827 Tool calls require user confirmation before execution.78287829 - `type: "always_ask"`78307831 - `BetaManagedAgentsAutoPolicy object`78327833 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.78347835 - `type: "auto"`78367837### Beta Managed Agents MCP Server URL Definition78387839- `BetaManagedAgentsMCPServerURLDefinition object`78407841 URL-based MCP server connection as returned in API responses.78427843 - `type: "url"`78447845 - `name: string`78467847 - `url: string`78487849### Beta Managed Agents MCP Tool Config78507851- `BetaManagedAgentsMCPToolConfig object`78527853 Resolved configuration for a specific MCP tool.78547855 - `enabled: boolean`78567857 - `name: string`78587859 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`78607861 Permission policy for tool execution.78627863 - `BetaManagedAgentsAlwaysAllowPolicy object`78647865 Tool calls are automatically approved without user confirmation.78667867 - `type: "always_allow"`78687869 - `BetaManagedAgentsAlwaysAskPolicy object`78707871 Tool calls require user confirmation before execution.78727873 - `type: "always_ask"`78747875 - `BetaManagedAgentsAutoPolicy object`78767877 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.78787879 - `type: "auto"`78807881### Beta Managed Agents MCP Tool Config Params78827883- `BetaManagedAgentsMCPToolConfigParams object`78847885 Configuration override for a specific MCP tool.78867887 - `name: string`78887889 Name of the MCP tool to configure. 1-128 characters.78907891 minLength: 1, maxLength: 12878927893 - `enabled: optional boolean or null`78947895 Whether this tool is enabled. Overrides the `default_config` setting.78967897 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`78987899 Permission policy for tool execution.79007901 - `BetaManagedAgentsAlwaysAllowPolicy object`79027903 Tool calls are automatically approved without user confirmation.79047905 - `type: "always_allow"`79067907 - `BetaManagedAgentsAlwaysAskPolicy object`79087909 Tool calls require user confirmation before execution.79107911 - `type: "always_ask"`79127913 - `BetaManagedAgentsAutoPolicy object`79147915 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.79167917 - `type: "auto"`79187919### Beta Managed Agents MCP Toolset79207921- `BetaManagedAgentsMCPToolset object`79227923 - `type: "mcp_toolset"`79247925 - `configs: array of BetaManagedAgentsMCPToolConfig`79267927 - `enabled: boolean`79287929 - `name: string`79307931 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`79327933 Permission policy for tool execution.79347935 - `BetaManagedAgentsAlwaysAllowPolicy object`79367937 Tool calls are automatically approved without user confirmation.79387939 - `type: "always_allow"`79407941 - `BetaManagedAgentsAlwaysAskPolicy object`79427943 Tool calls require user confirmation before execution.79447945 - `type: "always_ask"`79467947 - `BetaManagedAgentsAutoPolicy object`79487949 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.79507951 - `type: "auto"`79527953 - `default_config: BetaManagedAgentsMCPToolsetDefaultConfig`79547955 Resolved default configuration for all tools from an MCP server.79567957 - `enabled: boolean`79587959 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`79607961 Permission policy for tool execution.79627963 - `BetaManagedAgentsAlwaysAllowPolicy object`79647965 Tool calls are automatically approved without user confirmation.79667967 - `BetaManagedAgentsAlwaysAskPolicy object`79687969 Tool calls require user confirmation before execution.79707971 - `BetaManagedAgentsAutoPolicy object`79727973 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.79747975 - `mcp_server_name: string`79767977### Beta Managed Agents MCP Toolset Default Config79787979- `BetaManagedAgentsMCPToolsetDefaultConfig object`79807981 Resolved default configuration for all tools from an MCP server.79827983 - `enabled: boolean`79847985 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`79867987 Permission policy for tool execution.79887989 - `BetaManagedAgentsAlwaysAllowPolicy object`79907991 Tool calls are automatically approved without user confirmation.79927993 - `type: "always_allow"`79947995 - `BetaManagedAgentsAlwaysAskPolicy object`79967997 Tool calls require user confirmation before execution.79987999 - `type: "always_ask"`80008001 - `BetaManagedAgentsAutoPolicy object`80028003 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.80048005 - `type: "auto"`80068007### Beta Managed Agents MCP Toolset Default Config Params80088009- `BetaManagedAgentsMCPToolsetDefaultConfigParams object`80108011 Default configuration for all tools from an MCP server.80128013 - `enabled: optional boolean or null`80148015 Whether tools are enabled by default. Defaults to true if not specified.80168017 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`80188019 Permission policy for tool execution.80208021 - `BetaManagedAgentsAlwaysAllowPolicy object`80228023 Tool calls are automatically approved without user confirmation.80248025 - `type: "always_allow"`80268027 - `BetaManagedAgentsAlwaysAskPolicy object`80288029 Tool calls require user confirmation before execution.80308031 - `type: "always_ask"`80328033 - `BetaManagedAgentsAutoPolicy object`80348035 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.80368037 - `type: "auto"`80388039### Beta Managed Agents MCP Toolset Params80408041- `BetaManagedAgentsMCPToolsetParams object`80428043 Configuration for tools from an MCP server defined in `mcp_servers`.80448045 - `type: "mcp_toolset"`80468047 - `mcp_server_name: string`80488049 Name of the MCP server. Must match a server name from the mcp_servers array. 1-255 characters.80508051 minLength: 1, maxLength: 25580528053 - `configs: optional array of BetaManagedAgentsMCPToolConfigParams`80548055 Per-tool configuration overrides.80568057 - `name: string`80588059 Name of the MCP tool to configure. 1-128 characters.80608061 minLength: 1, maxLength: 12880628063 - `enabled: optional boolean or null`80648065 Whether this tool is enabled. Overrides the `default_config` setting.80668067 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`80688069 Permission policy for tool execution.80708071 - `BetaManagedAgentsAlwaysAllowPolicy object`80728073 Tool calls are automatically approved without user confirmation.80748075 - `type: "always_allow"`80768077 - `BetaManagedAgentsAlwaysAskPolicy object`80788079 Tool calls require user confirmation before execution.80808081 - `type: "always_ask"`80828083 - `BetaManagedAgentsAutoPolicy object`80848085 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.80868087 - `type: "auto"`80888089 - `default_config: optional BetaManagedAgentsMCPToolsetDefaultConfigParams or null`80908091 Default configuration for all tools from an MCP server.80928093 - `enabled: optional boolean or null`80948095 Whether tools are enabled by default. Defaults to true if not specified.80968097 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`80988099 Permission policy for tool execution.81008101 - `BetaManagedAgentsAlwaysAllowPolicy object`81028103 Tool calls are automatically approved without user confirmation.81048105 - `BetaManagedAgentsAlwaysAskPolicy object`81068107 Tool calls require user confirmation before execution.81088109 - `BetaManagedAgentsAutoPolicy object`81108111 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.81128113### Beta Managed Agents Model81148115- `BetaManagedAgentsModel = "claude-fable-5-1" or "claude-sonnet-5" or "claude-fable-5" or 11 more or string`81168117 The model that will power your agent.81188119 See [models](https://docs.anthropic.com/en/docs/models-overview) for additional details and options.81208121 - `"claude-fable-5-1" or "claude-sonnet-5" or "claude-fable-5" or 11 more`81228123 The model that will power your agent.81248125 See [models](https://docs.anthropic.com/en/docs/models-overview) for additional details and options.81268127 - `"claude-fable-5-1"`81288129 Frontier intelligence for ambitious tasks across coding, scientific discovery, and enterprise workflows81308131 - `"claude-sonnet-5"`81328133 High-performance model for coding and agents81348135 - `"claude-fable-5"`81368137 Next generation of intelligence for the hardest knowledge work and coding problems81388139 - `"claude-opus-5"`81408141 Powerful intelligence for long-running agents and coding81428143 - `"claude-opus-4-8"`81448145 Powerful intelligence for long-running agents and coding81468147 - `"claude-opus-4-7"`81488149 Powerful intelligence for long-running agents and coding81508151 - `"claude-opus-4-6"`81528153 Powerful intelligence for long-running agents and coding81548155 - `"claude-sonnet-4-6"`81568157 Best combination of speed and intelligence81588159 - `"claude-haiku-4-5"`81608161 Fastest model with near-frontier intelligence81628163 - `"claude-haiku-4-5-20251001"`81648165 Fastest model with near-frontier intelligence81668167 - `"claude-opus-4-5"`81688169 Powerful intelligence for long-running agents and coding81708171 - `"claude-opus-4-5-20251101"`81728173 Powerful intelligence for long-running agents and coding81748175 - `"claude-sonnet-4-5"`81768177 High-performance model for agents and coding81788179 - `"claude-sonnet-4-5-20250929"`81808181 High-performance model for agents and coding81828183 - `string`81848185### Beta Managed Agents Model Config81868187- `BetaManagedAgentsModelConfig object`81888189 Model identifier and configuration.81908191 - `id: BetaManagedAgentsModel`81928193 The model that will power your agent.81948195 See [models](https://docs.anthropic.com/en/docs/models-overview) for additional details and options.81968197 - `"claude-fable-5-1" or "claude-sonnet-5" or "claude-fable-5" or 11 more`81988199 The model that will power your agent.82008201 See [models](https://docs.anthropic.com/en/docs/models-overview) for additional details and options.82028203 - `"claude-fable-5-1"`82048205 Frontier intelligence for ambitious tasks across coding, scientific discovery, and enterprise workflows82068207 - `"claude-sonnet-5"`82088209 High-performance model for coding and agents82108211 - `"claude-fable-5"`82128213 Next generation of intelligence for the hardest knowledge work and coding problems82148215 - `"claude-opus-5"`82168217 Powerful intelligence for long-running agents and coding82188219 - `"claude-opus-4-8"`82208221 Powerful intelligence for long-running agents and coding82228223 - `"claude-opus-4-7"`82248225 Powerful intelligence for long-running agents and coding82268227 - `"claude-opus-4-6"`82288229 Powerful intelligence for long-running agents and coding82308231 - `"claude-sonnet-4-6"`82328233 Best combination of speed and intelligence82348235 - `"claude-haiku-4-5"`82368237 Fastest model with near-frontier intelligence82388239 - `"claude-haiku-4-5-20251001"`82408241 Fastest model with near-frontier intelligence82428243 - `"claude-opus-4-5"`82448245 Powerful intelligence for long-running agents and coding82468247 - `"claude-opus-4-5-20251101"`82488249 Powerful intelligence for long-running agents and coding82508251 - `"claude-sonnet-4-5"`82528253 High-performance model for agents and coding82548255 - `"claude-sonnet-4-5-20250929"`82568257 High-performance model for agents and coding82588259 - `string`82608261 - `effort: optional BetaManagedAgentsEffortLow or BetaManagedAgentsEffortMedium or BetaManagedAgentsEffortHigh or 2 more`82628263 How hard Claude works on each turn. Sets `output_config.effort` on every Messages call the session makes.82648265 - `BetaManagedAgentsEffortLow object`82668267 Low effort. Favors latency over reasoning depth.82688269 - `type: "low"`82708271 - `BetaManagedAgentsEffortMedium object`82728273 Medium effort. Balances latency and reasoning depth.82748275 - `type: "medium"`82768277 - `BetaManagedAgentsEffortHigh object`82788279 High effort. Favors reasoning depth.82808281 - `type: "high"`82828283 - `BetaManagedAgentsEffortXhigh object`82848285 Extra-high effort. Not all models accept this level.82868287 - `type: "xhigh"`82888289 - `BetaManagedAgentsEffortMax object`82908291 Maximum effort. Favors reasoning depth over latency.82928293 - `type: "max"`82948295 - `inference_geo: optional string`82968297 Geographic region for model inference. When unset, requests fall through to the workspace's default_inference_geo.82988299 - `speed: optional "standard" or "fast"`83008301 Inference speed mode. `fast` provides significantly faster output token generation at premium pricing. Not all models support `fast`; invalid combinations are rejected at create time.83028303 - `"standard"`83048305 - `"fast"`83068307### Beta Managed Agents Model Config Params83088309- `BetaManagedAgentsModelConfigParams object`83108311 An object that defines additional configuration control over model use83128313 - `id: BetaManagedAgentsModel`83148315 The model that will power your agent.83168317 See [models](https://docs.anthropic.com/en/docs/models-overview) for additional details and options.83188319 - `"claude-fable-5-1" or "claude-sonnet-5" or "claude-fable-5" or 11 more`83208321 The model that will power your agent.83228323 See [models](https://docs.anthropic.com/en/docs/models-overview) for additional details and options.83248325 - `"claude-fable-5-1"`83268327 Frontier intelligence for ambitious tasks across coding, scientific discovery, and enterprise workflows83288329 - `"claude-sonnet-5"`83308331 High-performance model for coding and agents83328333 - `"claude-fable-5"`83348335 Next generation of intelligence for the hardest knowledge work and coding problems83368337 - `"claude-opus-5"`83388339 Powerful intelligence for long-running agents and coding83408341 - `"claude-opus-4-8"`83428343 Powerful intelligence for long-running agents and coding83448345 - `"claude-opus-4-7"`83468347 Powerful intelligence for long-running agents and coding83488349 - `"claude-opus-4-6"`83508351 Powerful intelligence for long-running agents and coding83528353 - `"claude-sonnet-4-6"`83548355 Best combination of speed and intelligence83568357 - `"claude-haiku-4-5"`83588359 Fastest model with near-frontier intelligence83608361 - `"claude-haiku-4-5-20251001"`83628363 Fastest model with near-frontier intelligence83648365 - `"claude-opus-4-5"`83668367 Powerful intelligence for long-running agents and coding83688369 - `"claude-opus-4-5-20251101"`83708371 Powerful intelligence for long-running agents and coding83728373 - `"claude-sonnet-4-5"`83748375 High-performance model for agents and coding83768377 - `"claude-sonnet-4-5-20250929"`83788379 High-performance model for agents and coding83808381 - `string`83828383 - `effort: optional "low" or "medium" or "high" or 2 more or BetaManagedAgentsEffortLow or BetaManagedAgentsEffortMedium or 3 more or null`83848385 How hard Claude works on each inference call. Accepts a bare level string (`"high"`) or `{"type": "high"}`. On create, omitting it resolves the per-model default; on update, omitting it leaves the stored value unchanged.83868387 - `BetaManagedAgentsEffortLevel = "low" or "medium" or "high" or 2 more`83888389 How hard Claude works on each turn. Higher levels favor reasoning depth over latency. Not all models accept every level; invalid combinations are rejected at create time.83908391 - `"low"`83928393 - `"medium"`83948395 - `"high"`83968397 - `"xhigh"`83988399 - `"max"`84008401 - `BetaManagedAgentsEffortLow object`84028403 Low effort. Favors latency over reasoning depth.84048405 - `type: "low"`84068407 - `BetaManagedAgentsEffortMedium object`84088409 Medium effort. Balances latency and reasoning depth.84108411 - `type: "medium"`84128413 - `BetaManagedAgentsEffortHigh object`84148415 High effort. Favors reasoning depth.84168417 - `type: "high"`84188419 - `BetaManagedAgentsEffortXhigh object`84208421 Extra-high effort. Not all models accept this level.84228423 - `type: "xhigh"`84248425 - `BetaManagedAgentsEffortMax object`84268427 Maximum effort. Favors reasoning depth over latency.84288429 - `type: "max"`84308431 - `inference_geo: optional string or null`84328433 Geographic region for model inference. When unset, requests fall through to the workspace's default_inference_geo. On update, `model` is whole-object replacement — omitting inference_geo clears it.84348435 - `speed: optional "standard" or "fast" or null`84368437 Inference speed mode. `fast` provides significantly faster output token generation at premium pricing. Not all models support `fast`; invalid combinations are rejected at create time.84388439 - `"standard"`84408441 - `"fast"`84428443### Beta Managed Agents Multiagent Coordinator84448445- `BetaManagedAgentsMultiagentCoordinator object`84468447 Resolved coordinator topology with a concrete agent roster.84488449 - `type: "coordinator"`84508451 - `agents: array of BetaManagedAgentsAgentReference or BetaManagedAgentsAdvisor`84528453 Agents the coordinator may spawn as session threads, each resolved to a specific version.84548455 - `BetaManagedAgentsAgentReference object`84568457 A resolved agent reference with a concrete version.84588459 - `type: "agent"`84608461 - `id: string`84628463 - `version: number`84648465 format: int3284668467 - `BetaManagedAgentsAdvisor object`84688469 Platform advisor roster entry: a model the session's primary thread may consult mid-turn.84708471 - `type: "advisor"`84728473 - `model: string`84748475 The advisor model id.84768477### Beta Managed Agents Multiagent Coordinator Params84788479- `BetaManagedAgentsMultiagentCoordinatorParams object`84808481 A coordinator topology: the session's primary thread orchestrates work by spawning session threads, each running an agent drawn from the `agents` roster.84828483 - `type: "coordinator"`84848485 - `agents: array of BetaManagedAgentsMultiagentRosterEntryParams`84868487 Agents the coordinator may spawn as session threads. 1–20 entries. Each entry is an agent ID string, a versioned `{"type":"agent","id","version"}` reference, or `{"type":"self"}` to allow recursive self-invocation. Entries must reference distinct agents (after resolving `self` and string forms); at most one `self`. Referenced agents must exist, must not be archived, and must not themselves have `multiagent` set (depth limit 1).84888489 - `string`84908491 - `BetaManagedAgentsAgentParams object`84928493 Specification for an Agent. Provide a specific `version` or use the short-form `agent="agent_id"` for the most recent version84948495 - `type: "agent"`84968497 - `id: string`84988499 The `agent` ID.85008501 minLength: 1, maxLength: 12885028503 - `version: optional number`85048505 The specific `agent` version to use. Omit to use the latest version. Must be at least 1 if specified.85068507 format: int3285088509 - `BetaManagedAgentsMultiagentSelfParams object`85108511 Sentinel roster entry meaning "the agent that owns this configuration". Resolved server-side to a concrete agent reference.85128513 - `type: "self"`85148515 - `BetaManagedAgentsAdvisorParams object`85168517 Platform advisor roster entry: a model the session's primary thread may consult mid-turn. At most one per roster; the entry occupies the roster name `anthropic.advisor`.85188519 - `type: "advisor"`85208521 - `model: string`85228523 A Claude model id. The model must be permitted as an advisor for this agent's model — see the sessions/threads/advisor spec.85248525 minLength: 1, maxLength: 25685268527### Beta Managed Agents Multiagent Self Params85288529- `BetaManagedAgentsMultiagentSelfParams object`85308531 Sentinel roster entry meaning "the agent that owns this configuration". Resolved server-side to a concrete agent reference.85328533 - `type: "self"`85348535### Beta Managed Agents Read Tool Config85368537- `BetaManagedAgentsReadToolConfig object`85388539 Configuration for the read tool.85408541 - `type: "read"`85428543 - `enabled: boolean`85448545 - `name: "read"`85468547 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`85488549 Permission policy for tool execution.85508551 - `BetaManagedAgentsAlwaysAllowPolicy object`85528553 Tool calls are automatically approved without user confirmation.85548555 - `type: "always_allow"`85568557 - `BetaManagedAgentsAlwaysAskPolicy object`85588559 Tool calls require user confirmation before execution.85608561 - `type: "always_ask"`85628563 - `BetaManagedAgentsAutoPolicy object`85648565 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.85668567 - `type: "auto"`85688569### Beta Managed Agents Read Tool Config Params85708571- `BetaManagedAgentsReadToolConfigParams object`85728573 Configuration override for the read tool.85748575 - `type: optional "read"`85768577 - `name: "read"`85788579 Must be "read".85808581 - `enabled: optional boolean or null`85828583 Whether this tool is enabled and available to Claude. Overrides the default_config setting.85848585 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`85868587 Permission policy for tool execution.85888589 - `BetaManagedAgentsAlwaysAllowPolicy object`85908591 Tool calls are automatically approved without user confirmation.85928593 - `type: "always_allow"`85948595 - `BetaManagedAgentsAlwaysAskPolicy object`85968597 Tool calls require user confirmation before execution.85988599 - `type: "always_ask"`86008601 - `BetaManagedAgentsAutoPolicy object`86028603 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.86048605 - `type: "auto"`86068607### Beta Managed Agents Session Thread Agent86088609- `BetaManagedAgentsSessionThreadAgent object`86108611 Resolved `agent` definition for a single `session_thread`. Snapshot of the agent at thread creation time. The multiagent roster is not repeated here; read it from `Session.agent`.86128613 - `type: "agent"`86148615 - `id: string`86168617 - `description: string or null`86188619 - `mcp_servers: array of BetaManagedAgentsMCPServerURLDefinition`86208621 - `type: "url"`86228623 - `name: string`86248625 - `url: string`86268627 - `model: BetaManagedAgentsModelConfig`86288629 Model identifier and configuration.86308631 - `id: BetaManagedAgentsModel`86328633 The model that will power your agent.86348635 See [models](https://docs.anthropic.com/en/docs/models-overview) for additional details and options.86368637 - `"claude-fable-5-1" or "claude-sonnet-5" or "claude-fable-5" or 11 more`86388639 The model that will power your agent.86408641 See [models](https://docs.anthropic.com/en/docs/models-overview) for additional details and options.86428643 - `"claude-fable-5-1"`86448645 Frontier intelligence for ambitious tasks across coding, scientific discovery, and enterprise workflows86468647 - `"claude-sonnet-5"`86488649 High-performance model for coding and agents86508651 - `"claude-fable-5"`86528653 Next generation of intelligence for the hardest knowledge work and coding problems86548655 - `"claude-opus-5"`86568657 Powerful intelligence for long-running agents and coding86588659 - `"claude-opus-4-8"`86608661 Powerful intelligence for long-running agents and coding86628663 - `"claude-opus-4-7"`86648665 Powerful intelligence for long-running agents and coding86668667 - `"claude-opus-4-6"`86688669 Powerful intelligence for long-running agents and coding86708671 - `"claude-sonnet-4-6"`86728673 Best combination of speed and intelligence86748675 - `"claude-haiku-4-5"`86768677 Fastest model with near-frontier intelligence86788679 - `"claude-haiku-4-5-20251001"`86808681 Fastest model with near-frontier intelligence86828683 - `"claude-opus-4-5"`86848685 Powerful intelligence for long-running agents and coding86868687 - `"claude-opus-4-5-20251101"`86888689 Powerful intelligence for long-running agents and coding86908691 - `"claude-sonnet-4-5"`86928693 High-performance model for agents and coding86948695 - `"claude-sonnet-4-5-20250929"`86968697 High-performance model for agents and coding86988699 - `string`87008701 - `effort: optional BetaManagedAgentsEffortLow or BetaManagedAgentsEffortMedium or BetaManagedAgentsEffortHigh or 2 more`87028703 How hard Claude works on each turn. Sets `output_config.effort` on every Messages call the session makes.87048705 - `BetaManagedAgentsEffortLow object`87068707 Low effort. Favors latency over reasoning depth.87088709 - `type: "low"`87108711 - `BetaManagedAgentsEffortMedium object`87128713 Medium effort. Balances latency and reasoning depth.87148715 - `type: "medium"`87168717 - `BetaManagedAgentsEffortHigh object`87188719 High effort. Favors reasoning depth.87208721 - `type: "high"`87228723 - `BetaManagedAgentsEffortXhigh object`87248725 Extra-high effort. Not all models accept this level.87268727 - `type: "xhigh"`87288729 - `BetaManagedAgentsEffortMax object`87308731 Maximum effort. Favors reasoning depth over latency.87328733 - `type: "max"`87348735 - `inference_geo: optional string`87368737 Geographic region for model inference. When unset, requests fall through to the workspace's default_inference_geo.87388739 - `speed: optional "standard" or "fast"`87408741 Inference speed mode. `fast` provides significantly faster output token generation at premium pricing. Not all models support `fast`; invalid combinations are rejected at create time.87428743 - `"standard"`87448745 - `"fast"`87468747 - `name: string`87488749 - `skills: array of BetaManagedAgentsAnthropicSkill or BetaManagedAgentsCustomSkill`87508751 - `BetaManagedAgentsAnthropicSkill object`87528753 A resolved Anthropic-managed skill.87548755 - `type: "anthropic"`87568757 - `skill_id: string`87588759 - `version: string`87608761 - `BetaManagedAgentsCustomSkill object`87628763 A resolved user-created custom skill.87648765 - `type: "custom"`87668767 - `skill_id: string`87688769 - `version: string`87708771 - `system: string or null`87728773 - `tools: array of BetaManagedAgentsAgentToolset20260401 or BetaManagedAgentsMCPToolset or BetaManagedAgentsCustomTool`87748775 - `BetaManagedAgentsAgentToolset20260401 object`87768777 - `type: "agent_toolset_20260401"`87788779 - `configs: array of BetaManagedAgentsAgentToolConfig`87808781 - `BetaManagedAgentsBashToolConfig object`87828783 Configuration for the bash tool.87848785 - `type: "bash"`87868787 - `enabled: boolean`87888789 - `name: "bash"`87908791 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`87928793 Permission policy for tool execution.87948795 - `BetaManagedAgentsAlwaysAllowPolicy object`87968797 Tool calls are automatically approved without user confirmation.87988799 - `type: "always_allow"`88008801 - `BetaManagedAgentsAlwaysAskPolicy object`88028803 Tool calls require user confirmation before execution.88048805 - `type: "always_ask"`88068807 - `BetaManagedAgentsAutoPolicy object`88088809 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.88108811 - `type: "auto"`88128813 - `BetaManagedAgentsEditToolConfig object`88148815 Configuration for the edit tool.88168817 - `type: "edit"`88188819 - `enabled: boolean`88208821 - `name: "edit"`88228823 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`88248825 Permission policy for tool execution.88268827 - `BetaManagedAgentsAlwaysAllowPolicy object`88288829 Tool calls are automatically approved without user confirmation.88308831 - `BetaManagedAgentsAlwaysAskPolicy object`88328833 Tool calls require user confirmation before execution.88348835 - `BetaManagedAgentsAutoPolicy object`88368837 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.88388839 - `BetaManagedAgentsReadToolConfig object`88408841 Configuration for the read tool.88428843 - `type: "read"`88448845 - `enabled: boolean`88468847 - `name: "read"`88488849 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`88508851 Permission policy for tool execution.88528853 - `BetaManagedAgentsAlwaysAllowPolicy object`88548855 Tool calls are automatically approved without user confirmation.88568857 - `BetaManagedAgentsAlwaysAskPolicy object`88588859 Tool calls require user confirmation before execution.88608861 - `BetaManagedAgentsAutoPolicy object`88628863 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.88648865 - `BetaManagedAgentsWriteToolConfig object`88668867 Configuration for the write tool.88688869 - `type: "write"`88708871 - `enabled: boolean`88728873 - `name: "write"`88748875 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`88768877 Permission policy for tool execution.88788879 - `BetaManagedAgentsAlwaysAllowPolicy object`88808881 Tool calls are automatically approved without user confirmation.88828883 - `BetaManagedAgentsAlwaysAskPolicy object`88848885 Tool calls require user confirmation before execution.88868887 - `BetaManagedAgentsAutoPolicy object`88888889 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.88908891 - `BetaManagedAgentsGlobToolConfig object`88928893 Configuration for the glob tool.88948895 - `type: "glob"`88968897 - `enabled: boolean`88988899 - `name: "glob"`89008901 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`89028903 Permission policy for tool execution.89048905 - `BetaManagedAgentsAlwaysAllowPolicy object`89068907 Tool calls are automatically approved without user confirmation.89088909 - `BetaManagedAgentsAlwaysAskPolicy object`89108911 Tool calls require user confirmation before execution.89128913 - `BetaManagedAgentsAutoPolicy object`89148915 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.89168917 - `BetaManagedAgentsGrepToolConfig object`89188919 Configuration for the grep tool.89208921 - `type: "grep"`89228923 - `enabled: boolean`89248925 - `name: "grep"`89268927 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`89288929 Permission policy for tool execution.89308931 - `BetaManagedAgentsAlwaysAllowPolicy object`89328933 Tool calls are automatically approved without user confirmation.89348935 - `BetaManagedAgentsAlwaysAskPolicy object`89368937 Tool calls require user confirmation before execution.89388939 - `BetaManagedAgentsAutoPolicy object`89408941 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.89428943 - `BetaManagedAgentsWebFetchToolConfig object`89448945 Configuration for the web_fetch tool.89468947 - `type: "web_fetch"`89488949 - `enabled: boolean`89508951 - `name: "web_fetch"`89528953 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`89548955 Permission policy for tool execution.89568957 - `BetaManagedAgentsAlwaysAllowPolicy object`89588959 Tool calls are automatically approved without user confirmation.89608961 - `BetaManagedAgentsAlwaysAskPolicy object`89628963 Tool calls require user confirmation before execution.89648965 - `BetaManagedAgentsAutoPolicy object`89668967 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.89688969 - `allowed_domains: optional array of string`89708971 - `blocked_domains: optional array of string`89728973 - `max_content_tokens: optional number or null`89748975 format: int3289768977 - `BetaManagedAgentsWebSearchToolConfig object`89788979 Configuration for the web_search tool.89808981 - `type: "web_search"`89828983 - `enabled: boolean`89848985 - `name: "web_search"`89868987 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`89888989 Permission policy for tool execution.89908991 - `BetaManagedAgentsAlwaysAllowPolicy object`89928993 Tool calls are automatically approved without user confirmation.89948995 - `BetaManagedAgentsAlwaysAskPolicy object`89968997 Tool calls require user confirmation before execution.89988999 - `BetaManagedAgentsAutoPolicy object`90009001 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.90029003 - `allowed_domains: optional array of string`90049005 - `blocked_domains: optional array of string`90069007 - `user_location: optional BetaManagedAgentsUserLocation or null`90089009 Approximate user location for search result localization.90109011 - `type: "approximate"`90129013 Location precision. Only "approximate" is supported.90149015 - `city: optional string or null`90169017 City name.90189019 minLength: 1, maxLength: 25590209021 - `country: optional string or null`90229023 Two-letter ISO 3166-1 country code, uppercase.90249025 - `region: optional string or null`90269027 Region or state name.90289029 minLength: 1, maxLength: 25590309031 - `timezone: optional string or null`90329033 IANA timezone identifier, e.g. "America/Los_Angeles".90349035 minLength: 1, maxLength: 25590369037 - `default_config: BetaManagedAgentsAgentToolsetDefaultConfig`90389039 Resolved default configuration for agent tools.90409041 - `enabled: boolean`90429043 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`90449045 Permission policy for tool execution.90469047 - `BetaManagedAgentsAlwaysAllowPolicy object`90489049 Tool calls are automatically approved without user confirmation.90509051 - `BetaManagedAgentsAlwaysAskPolicy object`90529053 Tool calls require user confirmation before execution.90549055 - `BetaManagedAgentsAutoPolicy object`90569057 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.90589059 - `BetaManagedAgentsMCPToolset object`90609061 - `type: "mcp_toolset"`90629063 - `configs: array of BetaManagedAgentsMCPToolConfig`90649065 - `enabled: boolean`90669067 - `name: string`90689069 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`90709071 Permission policy for tool execution.90729073 - `BetaManagedAgentsAlwaysAllowPolicy object`90749075 Tool calls are automatically approved without user confirmation.90769077 - `BetaManagedAgentsAlwaysAskPolicy object`90789079 Tool calls require user confirmation before execution.90809081 - `BetaManagedAgentsAutoPolicy object`90829083 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.90849085 - `default_config: BetaManagedAgentsMCPToolsetDefaultConfig`90869087 Resolved default configuration for all tools from an MCP server.90889089 - `enabled: boolean`90909091 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`90929093 Permission policy for tool execution.90949095 - `BetaManagedAgentsAlwaysAllowPolicy object`90969097 Tool calls are automatically approved without user confirmation.90989099 - `BetaManagedAgentsAlwaysAskPolicy object`91009101 Tool calls require user confirmation before execution.91029103 - `BetaManagedAgentsAutoPolicy object`91049105 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.91069107 - `mcp_server_name: string`91089109 - `BetaManagedAgentsCustomTool object`91109111 A custom tool as returned in API responses.91129113 - `type: "custom"`91149115 - `description: string`91169117 - `input_schema: BetaManagedAgentsCustomToolInputSchema`91189119 JSON Schema for custom tool input parameters.91209121 - `type: "object"`91229123 - `properties: optional map[unknown] or null`91249125 - `required: optional array of string or null`91269127 - `name: string`91289129 - `version: number`91309131 format: int3291329133### Beta Managed Agents Skill Params91349135- `BetaManagedAgentsSkillParams = BetaManagedAgentsAnthropicSkillParams or BetaManagedAgentsCustomSkillParams`91369137 Skill to load in the session container.91389139 - `BetaManagedAgentsAnthropicSkillParams object`91409141 An Anthropic-managed skill.91429143 - `type: "anthropic"`91449145 - `skill_id: string`91469147 Identifier of the Anthropic skill (e.g., "xlsx").91489149 minLength: 1, maxLength: 6491509151 - `version: optional string or null`91529153 Version to pin. Defaults to latest if omitted.91549155 minLength: 1, maxLength: 6491569157 - `BetaManagedAgentsCustomSkillParams object`91589159 A user-created custom skill.91609161 - `type: "custom"`91629163 - `skill_id: string`91649165 Tagged ID of the custom skill (e.g., "skill_01XJ5...").91669167 minLength: 1, maxLength: 6491689169 - `version: optional string or null`91709171 Version to pin. Defaults to latest if omitted.91729173 minLength: 1, maxLength: 6491749175### Beta Managed Agents URL MCP Server Params91769177- `BetaManagedAgentsURLMCPServerParams object`91789179 URL-based MCP server connection.91809181 - `type: "url"`91829183 - `name: string`91849185 Unique name for this server, referenced by mcp_toolset configurations. 1-255 characters.91869187 minLength: 1, maxLength: 25591889189 - `url: string`91909191 Endpoint URL for the MCP server.91929193 maxLength: 204891949195### Beta Managed Agents User Location91969197- `BetaManagedAgentsUserLocation object`91989199 Approximate user location for search result localization.92009201 - `type: "approximate"`92029203 Location precision. Only "approximate" is supported.92049205 - `city: optional string or null`92069207 City name.92089209 minLength: 1, maxLength: 25592109211 - `country: optional string or null`92129213 Two-letter ISO 3166-1 country code, uppercase.92149215 - `region: optional string or null`92169217 Region or state name.92189219 minLength: 1, maxLength: 25592209221 - `timezone: optional string or null`92229223 IANA timezone identifier, e.g. "America/Los_Angeles".92249225 minLength: 1, maxLength: 25592269227### Beta Managed Agents Web Fetch Tool Config92289229- `BetaManagedAgentsWebFetchToolConfig object`92309231 Configuration for the web_fetch tool.92329233 - `type: "web_fetch"`92349235 - `enabled: boolean`92369237 - `name: "web_fetch"`92389239 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`92409241 Permission policy for tool execution.92429243 - `BetaManagedAgentsAlwaysAllowPolicy object`92449245 Tool calls are automatically approved without user confirmation.92469247 - `type: "always_allow"`92489249 - `BetaManagedAgentsAlwaysAskPolicy object`92509251 Tool calls require user confirmation before execution.92529253 - `type: "always_ask"`92549255 - `BetaManagedAgentsAutoPolicy object`92569257 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.92589259 - `type: "auto"`92609261 - `allowed_domains: optional array of string`92629263 - `blocked_domains: optional array of string`92649265 - `max_content_tokens: optional number or null`92669267 format: int3292689269### Beta Managed Agents Web Fetch Tool Config Params92709271- `BetaManagedAgentsWebFetchToolConfigParams object`92729273 Configuration override for the web_fetch tool.92749275 - `type: optional "web_fetch"`92769277 - `name: "web_fetch"`92789279 Must be "web_fetch".92809281 - `allowed_domains: optional array of string`92829283 Only fetch URLs whose host is one of these domains or a subdomain of one. Each entry is a plain hostname like "docs.example.com" (no scheme, port, or path). At most 64 entries; an empty list is rejected (omit the field instead). Cannot be combined with blocked_domains.92849285 - `blocked_domains: optional array of string`92869287 Never fetch URLs whose host is one of these domains or a subdomain of one. Each entry is a plain hostname like "ads.example.com" (no scheme, port, or path). At most 64 entries; an empty list is rejected (omit the field instead). Cannot be combined with allowed_domains.92889289 - `enabled: optional boolean or null`92909291 Whether this tool is enabled and available to Claude. Overrides the default_config setting.92929293 - `max_content_tokens: optional number or null`92949295 Maximum number of tokens of fetched text content to include in context per call. Does not apply to binary content such as PDFs.92969297 format: int3292989299 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`93009301 Permission policy for tool execution.93029303 - `BetaManagedAgentsAlwaysAllowPolicy object`93049305 Tool calls are automatically approved without user confirmation.93069307 - `type: "always_allow"`93089309 - `BetaManagedAgentsAlwaysAskPolicy object`93109311 Tool calls require user confirmation before execution.93129313 - `type: "always_ask"`93149315 - `BetaManagedAgentsAutoPolicy object`93169317 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.93189319 - `type: "auto"`93209321### Beta Managed Agents Web Search Tool Config93229323- `BetaManagedAgentsWebSearchToolConfig object`93249325 Configuration for the web_search tool.93269327 - `type: "web_search"`93289329 - `enabled: boolean`93309331 - `name: "web_search"`93329333 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`93349335 Permission policy for tool execution.93369337 - `BetaManagedAgentsAlwaysAllowPolicy object`93389339 Tool calls are automatically approved without user confirmation.93409341 - `type: "always_allow"`93429343 - `BetaManagedAgentsAlwaysAskPolicy object`93449345 Tool calls require user confirmation before execution.93469347 - `type: "always_ask"`93489349 - `BetaManagedAgentsAutoPolicy object`93509351 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.93529353 - `type: "auto"`93549355 - `allowed_domains: optional array of string`93569357 - `blocked_domains: optional array of string`93589359 - `user_location: optional BetaManagedAgentsUserLocation or null`93609361 Approximate user location for search result localization.93629363 - `type: "approximate"`93649365 Location precision. Only "approximate" is supported.93669367 - `city: optional string or null`93689369 City name.93709371 minLength: 1, maxLength: 25593729373 - `country: optional string or null`93749375 Two-letter ISO 3166-1 country code, uppercase.93769377 - `region: optional string or null`93789379 Region or state name.93809381 minLength: 1, maxLength: 25593829383 - `timezone: optional string or null`93849385 IANA timezone identifier, e.g. "America/Los_Angeles".93869387 minLength: 1, maxLength: 25593889389### Beta Managed Agents Web Search Tool Config Params93909391- `BetaManagedAgentsWebSearchToolConfigParams object`93929393 Configuration override for the web_search tool.93949395 - `type: optional "web_search"`93969397 - `name: "web_search"`93989399 Must be "web_search".94009401 - `allowed_domains: optional array of string`94029403 Only return search results whose host is one of these domains or a subdomain of one. Each entry is a plain hostname like "docs.example.com" (no scheme or port; an optional path suffix is accepted). At most 64 entries; an empty list is rejected (omit the field instead). Cannot be combined with blocked_domains.94049405 - `blocked_domains: optional array of string`94069407 Never return search results whose host is one of these domains or a subdomain of one. Each entry is a plain hostname like "ads.example.com" (no scheme or port; an optional path suffix is accepted). At most 64 entries; an empty list is rejected (omit the field instead). Cannot be combined with allowed_domains.94089409 - `enabled: optional boolean or null`94109411 Whether this tool is enabled and available to Claude. Overrides the default_config setting.94129413 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`94149415 Permission policy for tool execution.94169417 - `BetaManagedAgentsAlwaysAllowPolicy object`94189419 Tool calls are automatically approved without user confirmation.94209421 - `type: "always_allow"`94229423 - `BetaManagedAgentsAlwaysAskPolicy object`94249425 Tool calls require user confirmation before execution.94269427 - `type: "always_ask"`94289429 - `BetaManagedAgentsAutoPolicy object`94309431 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.94329433 - `type: "auto"`94349435 - `user_location: optional BetaManagedAgentsUserLocation or null`94369437 Approximate user location for search result localization.94389439 - `type: "approximate"`94409441 Location precision. Only "approximate" is supported.94429443 - `city: optional string or null`94449445 City name.94469447 minLength: 1, maxLength: 25594489449 - `country: optional string or null`94509451 Two-letter ISO 3166-1 country code, uppercase.94529453 - `region: optional string or null`94549455 Region or state name.94569457 minLength: 1, maxLength: 25594589459 - `timezone: optional string or null`94609461 IANA timezone identifier, e.g. "America/Los_Angeles".94629463 minLength: 1, maxLength: 25594649465### Beta Managed Agents Write Tool Config94669467- `BetaManagedAgentsWriteToolConfig object`94689469 Configuration for the write tool.94709471 - `type: "write"`94729473 - `enabled: boolean`94749475 - `name: "write"`94769477 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`94789479 Permission policy for tool execution.94809481 - `BetaManagedAgentsAlwaysAllowPolicy object`94829483 Tool calls are automatically approved without user confirmation.94849485 - `type: "always_allow"`94869487 - `BetaManagedAgentsAlwaysAskPolicy object`94889489 Tool calls require user confirmation before execution.94909491 - `type: "always_ask"`94929493 - `BetaManagedAgentsAutoPolicy object`94949495 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.94969497 - `type: "auto"`94989499### Beta Managed Agents Write Tool Config Params95009501- `BetaManagedAgentsWriteToolConfigParams object`95029503 Configuration override for the write tool.95049505 - `type: optional "write"`95069507 - `name: "write"`95089509 Must be "write".95109511 - `enabled: optional boolean or null`95129513 Whether this tool is enabled and available to Claude. Overrides the default_config setting.95149515 - `permission_policy: optional BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy or null`95169517 Permission policy for tool execution.95189519 - `BetaManagedAgentsAlwaysAllowPolicy object`95209521 Tool calls are automatically approved without user confirmation.95229523 - `type: "always_allow"`95249525 - `BetaManagedAgentsAlwaysAskPolicy object`95269527 Tool calls require user confirmation before execution.95289529 - `type: "always_ask"`95309531 - `BetaManagedAgentsAutoPolicy object`95329533 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.95349535 - `type: "auto"`95369537## Agents › Versions95389539### List Agent Versions95409541**GET** `/v1/agents/{agent_id}/versions`95429543List Agent Versions95449545#### Path parameters95469547- `agent_id: string`95489549#### Query parameters95509551- `limit: optional number`95529553 Maximum results per page. Default 20, maximum 100.95549555 format: int3295569557- `page: optional string`95589559 Opaque pagination cursor.95609561#### Headers95629563- `"anthropic-beta": optional array of AnthropicBeta`95649565 Optional header to specify the beta version(s) you want to use.95669567 - `string`95689569 - `"message-batches-2024-09-24" or "prompt-caching-2024-07-31" or "computer-use-2024-10-22" or 43 more`95709571 - `"message-batches-2024-09-24"`95729573 - `"prompt-caching-2024-07-31"`95749575 - `"computer-use-2024-10-22"`95769577 - `"computer-use-2025-01-24"`95789579 - `"pdfs-2024-09-25"`95809581 - `"token-counting-2024-11-01"`95829583 - `"token-efficient-tools-2025-02-19"`95849585 - `"output-128k-2025-02-19"`95869587 - `"files-api-2025-04-14"`95889589 - `"mcp-client-2025-04-04"`95909591 - `"mcp-client-2025-11-20"`95929593 - `"dev-full-thinking-2025-05-14"`95949595 - `"interleaved-thinking-2025-05-14"`95969597 - `"code-execution-2025-05-22"`95989599 - `"extended-cache-ttl-2025-04-11"`96009601 - `"context-1m-2025-08-07"`96029603 - `"context-management-2025-06-27"`96049605 - `"model-context-window-exceeded-2025-08-26"`96069607 - `"skills-2025-10-02"`96089609 - `"fast-mode-2026-02-01"`96109611 - `"output-300k-2026-03-24"`96129613 - `"user-profiles-2026-03-24"`96149615 - `"user-profiles-2026-08-18"`96169617 - `"user-profiles-2026-09-04"`96189619 - `"advisor-tool-2026-03-01"`96209621 - `"managed-agents-2026-04-01"`96229623 - `"cache-diagnosis-2026-04-07"`96249625 - `"dreaming-2026-04-21"`96269627 - `"thinking-token-count-2026-05-13"`96289629 - `"server-side-fallback-2026-06-01"`96309631 - `"server-side-fallback-2026-07-01"`96329633 - `"fallback-credit-2026-06-01"`96349635 - `"fallback-credit-2026-07-01"`96369637 - `"agent-memory-2026-07-22"`96389639 - `"mid-conversation-tool-changes-2026-07-01"`96409641 - `"compact-2026-01-12"`96429643 - `"computer-use-2025-11-24"`96449645 - `"mcp-tunnels-2026-06-22"`96469647 - `"structured-outputs-2025-11-13"`96489649 - `"task-budgets-2026-03-13"`96509651 - `"thinking-display-updates-2026-08-18"`96529653 - `"ce-user-management-2026-07-13"`96549655 - `"mid-conversation-output-config-2026-07-01"`96569657 - `"thinking-binding-controls-2026-08-01"`96589659 - `"mid-conversation-system-clear-at-2026-08-21"`96609661 - `"compact-2026-09-04"`96629663- `"anthropic-workspace-id": optional string`96649665#### Returns96669667- `data: array of BetaManagedAgentsAgent`96689669 Agent versions.96709671 - `type: "agent"`96729673 - `id: string`96749675 - `archived_at: string or null`96769677 A timestamp in RFC 3339 format96789679 format: date-time96809681 - `created_at: string`96829683 A timestamp in RFC 3339 format96849685 format: date-time96869687 - `description: string or null`96889689 - `mcp_servers: array of BetaManagedAgentsMCPServerURLDefinition`96909691 - `type: "url"`96929693 - `name: string`96949695 - `url: string`96969697 - `metadata: map[string]`96989699 - `model: BetaManagedAgentsModelConfig`97009701 Model identifier and configuration.97029703 - `id: BetaManagedAgentsModel`97049705 The model that will power your agent.97069707 See [models](https://docs.anthropic.com/en/docs/models-overview) for additional details and options.97089709 - `"claude-fable-5-1" or "claude-sonnet-5" or "claude-fable-5" or 11 more`97109711 The model that will power your agent.97129713 See [models](https://docs.anthropic.com/en/docs/models-overview) for additional details and options.97149715 - `"claude-fable-5-1"`97169717 Frontier intelligence for ambitious tasks across coding, scientific discovery, and enterprise workflows97189719 - `"claude-sonnet-5"`97209721 High-performance model for coding and agents97229723 - `"claude-fable-5"`97249725 Next generation of intelligence for the hardest knowledge work and coding problems97269727 - `"claude-opus-5"`97289729 Powerful intelligence for long-running agents and coding97309731 - `"claude-opus-4-8"`97329733 Powerful intelligence for long-running agents and coding97349735 - `"claude-opus-4-7"`97369737 Powerful intelligence for long-running agents and coding97389739 - `"claude-opus-4-6"`97409741 Powerful intelligence for long-running agents and coding97429743 - `"claude-sonnet-4-6"`97449745 Best combination of speed and intelligence97469747 - `"claude-haiku-4-5"`97489749 Fastest model with near-frontier intelligence97509751 - `"claude-haiku-4-5-20251001"`97529753 Fastest model with near-frontier intelligence97549755 - `"claude-opus-4-5"`97569757 Powerful intelligence for long-running agents and coding97589759 - `"claude-opus-4-5-20251101"`97609761 Powerful intelligence for long-running agents and coding97629763 - `"claude-sonnet-4-5"`97649765 High-performance model for agents and coding97669767 - `"claude-sonnet-4-5-20250929"`97689769 High-performance model for agents and coding97709771 - `string`97729773 - `effort: optional BetaManagedAgentsEffortLow or BetaManagedAgentsEffortMedium or BetaManagedAgentsEffortHigh or 2 more`97749775 How hard Claude works on each turn. Sets `output_config.effort` on every Messages call the session makes.97769777 - `BetaManagedAgentsEffortLow object`97789779 Low effort. Favors latency over reasoning depth.97809781 - `type: "low"`97829783 - `BetaManagedAgentsEffortMedium object`97849785 Medium effort. Balances latency and reasoning depth.97869787 - `type: "medium"`97889789 - `BetaManagedAgentsEffortHigh object`97909791 High effort. Favors reasoning depth.97929793 - `type: "high"`97949795 - `BetaManagedAgentsEffortXhigh object`97969797 Extra-high effort. Not all models accept this level.97989799 - `type: "xhigh"`98009801 - `BetaManagedAgentsEffortMax object`98029803 Maximum effort. Favors reasoning depth over latency.98049805 - `type: "max"`98069807 - `inference_geo: optional string`98089809 Geographic region for model inference. When unset, requests fall through to the workspace's default_inference_geo.98109811 - `speed: optional "standard" or "fast"`98129813 Inference speed mode. `fast` provides significantly faster output token generation at premium pricing. Not all models support `fast`; invalid combinations are rejected at create time.98149815 - `"standard"`98169817 - `"fast"`98189819 - `multiagent: BetaManagedAgentsMultiagent or null`98209821 Resolved coordinator topology with a concrete agent roster.98229823 - `type: "coordinator"`98249825 - `agents: array of BetaManagedAgentsAgentReference or BetaManagedAgentsAdvisor`98269827 Agents the coordinator may spawn as session threads, each resolved to a specific version.98289829 - `BetaManagedAgentsAgentReference object`98309831 A resolved agent reference with a concrete version.98329833 - `type: "agent"`98349835 - `id: string`98369837 - `version: number`98389839 format: int3298409841 - `BetaManagedAgentsAdvisor object`98429843 Platform advisor roster entry: a model the session's primary thread may consult mid-turn.98449845 - `type: "advisor"`98469847 - `model: string`98489849 The advisor model id.98509851 - `name: string`98529853 - `skills: array of BetaManagedAgentsAnthropicSkill or BetaManagedAgentsCustomSkill`98549855 - `BetaManagedAgentsAnthropicSkill object`98569857 A resolved Anthropic-managed skill.98589859 - `type: "anthropic"`98609861 - `skill_id: string`98629863 - `version: string`98649865 - `BetaManagedAgentsCustomSkill object`98669867 A resolved user-created custom skill.98689869 - `type: "custom"`98709871 - `skill_id: string`98729873 - `version: string`98749875 - `system: string or null`98769877 - `tools: array of BetaManagedAgentsAgentToolset20260401 or BetaManagedAgentsMCPToolset or BetaManagedAgentsCustomTool`98789879 - `BetaManagedAgentsAgentToolset20260401 object`98809881 - `type: "agent_toolset_20260401"`98829883 - `configs: array of BetaManagedAgentsAgentToolConfig`98849885 - `BetaManagedAgentsBashToolConfig object`98869887 Configuration for the bash tool.98889889 - `type: "bash"`98909891 - `enabled: boolean`98929893 - `name: "bash"`98949895 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`98969897 Permission policy for tool execution.98989899 - `BetaManagedAgentsAlwaysAllowPolicy object`99009901 Tool calls are automatically approved without user confirmation.99029903 - `type: "always_allow"`99049905 - `BetaManagedAgentsAlwaysAskPolicy object`99069907 Tool calls require user confirmation before execution.99089909 - `type: "always_ask"`99109911 - `BetaManagedAgentsAutoPolicy object`99129913 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.99149915 - `type: "auto"`99169917 - `BetaManagedAgentsEditToolConfig object`99189919 Configuration for the edit tool.99209921 - `type: "edit"`99229923 - `enabled: boolean`99249925 - `name: "edit"`99269927 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`99289929 Permission policy for tool execution.99309931 - `BetaManagedAgentsAlwaysAllowPolicy object`99329933 Tool calls are automatically approved without user confirmation.99349935 - `BetaManagedAgentsAlwaysAskPolicy object`99369937 Tool calls require user confirmation before execution.99389939 - `BetaManagedAgentsAutoPolicy object`99409941 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.99429943 - `BetaManagedAgentsReadToolConfig object`99449945 Configuration for the read tool.99469947 - `type: "read"`99489949 - `enabled: boolean`99509951 - `name: "read"`99529953 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`99549955 Permission policy for tool execution.99569957 - `BetaManagedAgentsAlwaysAllowPolicy object`99589959 Tool calls are automatically approved without user confirmation.99609961 - `BetaManagedAgentsAlwaysAskPolicy object`99629963 Tool calls require user confirmation before execution.99649965 - `BetaManagedAgentsAutoPolicy object`99669967 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.99689969 - `BetaManagedAgentsWriteToolConfig object`99709971 Configuration for the write tool.99729973 - `type: "write"`99749975 - `enabled: boolean`99769977 - `name: "write"`99789979 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`99809981 Permission policy for tool execution.99829983 - `BetaManagedAgentsAlwaysAllowPolicy object`99849985 Tool calls are automatically approved without user confirmation.99869987 - `BetaManagedAgentsAlwaysAskPolicy object`99889989 Tool calls require user confirmation before execution.99909991 - `BetaManagedAgentsAutoPolicy object`99929993 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.99949995 - `BetaManagedAgentsGlobToolConfig object`99969997 Configuration for the glob tool.99989999 - `type: "glob"`1000010001 - `enabled: boolean`1000210003 - `name: "glob"`1000410005 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`1000610007 Permission policy for tool execution.1000810009 - `BetaManagedAgentsAlwaysAllowPolicy object`1001010011 Tool calls are automatically approved without user confirmation.1001210013 - `BetaManagedAgentsAlwaysAskPolicy object`1001410015 Tool calls require user confirmation before execution.1001610017 - `BetaManagedAgentsAutoPolicy object`1001810019 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.1002010021 - `BetaManagedAgentsGrepToolConfig object`1002210023 Configuration for the grep tool.1002410025 - `type: "grep"`1002610027 - `enabled: boolean`1002810029 - `name: "grep"`1003010031 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`1003210033 Permission policy for tool execution.1003410035 - `BetaManagedAgentsAlwaysAllowPolicy object`1003610037 Tool calls are automatically approved without user confirmation.1003810039 - `BetaManagedAgentsAlwaysAskPolicy object`1004010041 Tool calls require user confirmation before execution.1004210043 - `BetaManagedAgentsAutoPolicy object`1004410045 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.1004610047 - `BetaManagedAgentsWebFetchToolConfig object`1004810049 Configuration for the web_fetch tool.1005010051 - `type: "web_fetch"`1005210053 - `enabled: boolean`1005410055 - `name: "web_fetch"`1005610057 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`1005810059 Permission policy for tool execution.1006010061 - `BetaManagedAgentsAlwaysAllowPolicy object`1006210063 Tool calls are automatically approved without user confirmation.1006410065 - `BetaManagedAgentsAlwaysAskPolicy object`1006610067 Tool calls require user confirmation before execution.1006810069 - `BetaManagedAgentsAutoPolicy object`1007010071 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.1007210073 - `allowed_domains: optional array of string`1007410075 - `blocked_domains: optional array of string`1007610077 - `max_content_tokens: optional number or null`1007810079 format: int321008010081 - `BetaManagedAgentsWebSearchToolConfig object`1008210083 Configuration for the web_search tool.1008410085 - `type: "web_search"`1008610087 - `enabled: boolean`1008810089 - `name: "web_search"`1009010091 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`1009210093 Permission policy for tool execution.1009410095 - `BetaManagedAgentsAlwaysAllowPolicy object`1009610097 Tool calls are automatically approved without user confirmation.1009810099 - `BetaManagedAgentsAlwaysAskPolicy object`1010010101 Tool calls require user confirmation before execution.1010210103 - `BetaManagedAgentsAutoPolicy object`1010410105 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.1010610107 - `allowed_domains: optional array of string`1010810109 - `blocked_domains: optional array of string`1011010111 - `user_location: optional BetaManagedAgentsUserLocation or null`1011210113 Approximate user location for search result localization.1011410115 - `type: "approximate"`1011610117 Location precision. Only "approximate" is supported.1011810119 - `city: optional string or null`1012010121 City name.1012210123 minLength: 1, maxLength: 2551012410125 - `country: optional string or null`1012610127 Two-letter ISO 3166-1 country code, uppercase.1012810129 - `region: optional string or null`1013010131 Region or state name.1013210133 minLength: 1, maxLength: 2551013410135 - `timezone: optional string or null`1013610137 IANA timezone identifier, e.g. "America/Los_Angeles".1013810139 minLength: 1, maxLength: 2551014010141 - `default_config: BetaManagedAgentsAgentToolsetDefaultConfig`1014210143 Resolved default configuration for agent tools.1014410145 - `enabled: boolean`1014610147 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`1014810149 Permission policy for tool execution.1015010151 - `BetaManagedAgentsAlwaysAllowPolicy object`1015210153 Tool calls are automatically approved without user confirmation.1015410155 - `BetaManagedAgentsAlwaysAskPolicy object`1015610157 Tool calls require user confirmation before execution.1015810159 - `BetaManagedAgentsAutoPolicy object`1016010161 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.1016210163 - `BetaManagedAgentsMCPToolset object`1016410165 - `type: "mcp_toolset"`1016610167 - `configs: array of BetaManagedAgentsMCPToolConfig`1016810169 - `enabled: boolean`1017010171 - `name: string`1017210173 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`1017410175 Permission policy for tool execution.1017610177 - `BetaManagedAgentsAlwaysAllowPolicy object`1017810179 Tool calls are automatically approved without user confirmation.1018010181 - `BetaManagedAgentsAlwaysAskPolicy object`1018210183 Tool calls require user confirmation before execution.1018410185 - `BetaManagedAgentsAutoPolicy object`1018610187 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.1018810189 - `default_config: BetaManagedAgentsMCPToolsetDefaultConfig`1019010191 Resolved default configuration for all tools from an MCP server.1019210193 - `enabled: boolean`1019410195 - `permission_policy: BetaManagedAgentsAlwaysAllowPolicy or BetaManagedAgentsAlwaysAskPolicy or BetaManagedAgentsAutoPolicy`1019610197 Permission policy for tool execution.1019810199 - `BetaManagedAgentsAlwaysAllowPolicy object`1020010201 Tool calls are automatically approved without user confirmation.1020210203 - `BetaManagedAgentsAlwaysAskPolicy object`1020410205 Tool calls require user confirmation before execution.1020610207 - `BetaManagedAgentsAutoPolicy object`1020810209 The server decides each tool call individually: it judges, from the tool, its input, and the session content so far, whether the call is safe to execute or high-risk, and evaluates it to allow when judged safe and to deny when judged high-risk. A call the server cannot reach a judgement on evaluates to ask.1021010211 - `mcp_server_name: string`1021210213 - `BetaManagedAgentsCustomTool object`1021410215 A custom tool as returned in API responses.1021610217 - `type: "custom"`1021810219 - `description: string`1022010221 - `input_schema: BetaManagedAgentsCustomToolInputSchema`1022210223 JSON Schema for custom tool input parameters.1022410225 - `type: "object"`1022610227 - `properties: optional map[unknown] or null`1022810229 - `required: optional array of string or null`1023010231 - `name: string`1023210233 - `updated_at: string`1023410235 A timestamp in RFC 3339 format1023610237 format: date-time1023810239 - `version: number`1024010241 The agent's current version. Starts at 1 and increments when the agent is modified.1024210243 format: int321024410245- `next_page: optional string or null`1024610247 Opaque cursor for the next page. Null when no more results.1024810249#### Example1025010251```bash10252curl https://api.anthropic.com/v1/agents/$AGENT_ID/versions \10253 -H 'anthropic-version: 2023-06-01' \10254 -H 'anthropic-beta: managed-agents-2026-04-01' \10255 -H "X-Api-Key: $ANTHROPIC_API_KEY"10256```1025710258##### Response (200)1025910260```json10261{10262 "data": [10263 {10264 "id": "agent_011CZkYpogX7uDKUyvBTophP",10265 "archived_at": null,10266 "created_at": "2026-03-15T10:00:00Z",10267 "description": "A general-purpose starter agent.",10268 "mcp_servers": [10269 {10270 "name": "example-mcp",10271 "type": "url",10272 "url": "https://example-server.modelcontextprotocol.io/sse"10273 }10274 ],10275 "metadata": {10276 "foo": "bar"10277 },10278 "model": {10279 "id": "claude-opus-5",10280 "effort": {10281 "type": "low"10282 },10283 "inference_geo": "inference_geo",10284 "speed": "standard"10285 },10286 "multiagent": {10287 "agents": [10288 {10289 "id": "agent_011CZkYqphY8vELVzwCUpqiQ",10290 "type": "agent",10291 "version": 110292 }10293 ],10294 "type": "coordinator"10295 },10296 "name": "My First Agent",10297 "skills": [10298 {10299 "skill_id": "xlsx",10300 "type": "anthropic",10301 "version": "1"10302 },10303 {10304 "skill_id": "skill_011CZkZFNu9hAbo3jZPRgTlx",10305 "type": "custom",10306 "version": "2"10307 }10308 ],10309 "system": "You are a general-purpose agent that can research, write code, run commands, and use connected tools to complete the user's task end to end.",10310 "tools": [10311 {10312 "configs": [10313 {10314 "enabled": true,10315 "name": "bash",10316 "permission_policy": {10317 "type": "always_allow"10318 },10319 "type": "bash"10320 }10321 ],10322 "default_config": {10323 "enabled": true,10324 "permission_policy": {10325 "type": "always_ask"10326 }10327 },10328 "type": "agent_toolset_20260401"10329 }10330 ],10331 "type": "agent",10332 "updated_at": "2026-03-15T10:00:00Z",10333 "version": 110334 }10335 ],10336 "next_page": "next_page"10337}10338```10339