# Anthropic `ant` CLI

Status: DOCUMENTED (not installed locally; no live run). Reference page api/cli/messages/create mirrors POST /v1/messages 1:1. Sources: https://platform.claude.com/docs/en/cli-sdks-libraries/cli/quickstart · …/cli/using · …/cli/authentication · …/cli/scripting · …/cli/apply · …/cli/sessions-connect · https://platform.claude.com/docs/en/api/cli/messages/create · https://platform.claude.com/docs/en/api/beta-headers Machine-readable: generated/fragments/sdks/anthropic-sdks.json (record language: cli) Last verified: 2026-09-18 (docs only)

# Install & auth

	Command
Homebrew	`brew install anthropics/tap/ant`
Linux tarball	`curl -fsSL https://github.com/anthropics/anthropic-cli/releases/download/v1.33.0/ant_1.33.0_${OS}_${ARCH}.tar.gz \| sudo tar -xz -C /usr/local/bin ant`
Go	`go install github.com/anthropics/anthropic-cli/cmd/ant@latest` (Go ≥1.25)
Check	`ant --version`
Login	`ant auth login` — browser OAuth against the Console, token scoped to one workspace, stored under `$ANTHROPIC_CONFIG_DIR/credentials/<profile>.json`
Admin	`ant auth login --scope org:admin` under a dedicated profile (org-wide)
API key	`ANTHROPIC_API_KEY` env (overrides every profile) or `--api-key`; `ANTHROPIC_AUTH_TOKEN` also honored; `ANTHROPIC_BASE_URL` / `--base-url`
Status / logout	`ant auth status` (which credential source & workspace won; not a health check) · `ant auth logout [--all]`
Profiles	`ant profile …` (`workspace_id`, `base_url`, `organization_id`, `scope`, `client_id`, `console_url`); `--profile` / `ANTHROPIC_PROFILE`; WIF for CI
Reuse token in curl	`ant auth print-credentials --access-token` → `Authorization: Bearer …` (keep `ANTHROPIC_API_KEY`/`ANTHROPIC_AUTH_TOKEN` unset)
Completion	`ant @completion zsh\|bash\|fish\|powershell`

# Command structure

ant <resource> <action>; nested resources use colons (ant beta:sessions:events stream). Beta resources live under beta: and add the right anthropic-beta header automatically; --beta a,b (or repeated) overrides/adds.

# Global flags

Flag	Meaning
`--format auto\|json\|jsonl\|yaml\|pretty\|raw\|explore`	`auto` = pretty JSON for create/modify; list/retrieve open the interactive explorer on a TTY, pretty JSON when piped. `jsonl` = one item/event per line. `raw` = raw body bytes, no auto-pagination
`--transform <GJSON path>`	reshape/filter (per item on list endpoints)
`-r`, `--raw-output`	unquoted strings (like `jq -r`) — not applied to errors
`--format-error`, `--transform-error`	same for error responses
`--workspace-id wrkspc_…`	`anthropic-workspace-id` header (`ANTHROPIC_WORKSPACE_ID`)
`--profile`, `--base-url`, `--api-key`, `--beta`	see above
`--debug`	full HTTP request/response on stderr, keys redacted (shows `request-id`)

# Request bodies

Scalars → flags; structured values accept relaxed YAML-ish or strict JSON (--message '{role: user, content: "Hello"}'); repeatable flags build arrays (--tool, --message, --event); stdin JSON/YAML document (heredoc <<'YAML') merged with flags (flags win); @path inlines file contents into string fields (binary auto-base64; @file:// text, @data:// base64; \@ literal).

# `ant messages create` (POST /v1/messages)

bash

ant messages create --model claude-opus-5 --max-tokens 1024 --message '{role: user, content: "Hello, Claude"}'
ant messages create --stream --format jsonl --model claude-opus-5 --max-tokens 1024 --message '{role: user, content: "Hello"}' \
  | jq -rj 'select(.delta.type? == "text_delta") | .delta.text'
ant --debug messages create --model claude-sonnet-5 --max-tokens 1024 --message '{role: user, content: "Hi"}'   # request-id on stderr
ant messages create --stream --format jsonl <<'YAML'
model: claude-sonnet-5
max_tokens: 128000
messages:
  - role: user
    content: Write a detailed analysis...
YAML

Flag	Body / header field
`--model`, `--max-tokens`, `--message` (repeatable), `--system` (string or text blocks; `@file.md`), `--stop-sequence` (repeatable), `--stream`	`model`, `max_tokens`, `messages[]`, `system`, `stop_sequences[]`, `stream`
`--metadata`, `--service-tier auto\|standard_only`, `--cache-control`, `--container`, `--inference-geo`, `--output-config`, `--thinking`, `--tool` (repeatable), `--tool-choice`	same-named body fields
`--temperature`, `--top-p`, `--top-k`	deprecated sampling params (same model rules as the API)
`--user-profile-id`, `--workspace-id`	`anthropic-user-profile-id` (beta), `anthropic-workspace-id` headers
`--beta a,b`	`anthropic-beta`

Output = the full Message JSON (or, with --stream --format jsonl, one SSE event object per line: message_start, content_block_start, content_block_delta{delta.type: text_delta…}, …). Sibling commands follow the same resource pattern: ant messages count-tokens, ant models list, ant models retrieve <id>, ant messages:batches …, ant files …, ant skills …, ant beta:agents|sessions|environments ….

# Scripting patterns

Chain ids: ant beta:agents list --transform id --raw-output | head -1 | xargs -I{} ant beta:agents retrieve --agent-id {}
Errors: ant … --format-error yaml --transform-error error.message
Claude Code can shell out to ant directly (no integration code).
ant apply [--dry-run|--yes|--force|--prune|--upgrade|--lock-file <p>|-v] — Managed Agents resources as files (agents/*.md, environments/*.yaml, memory_stores/, deployments/*.md, skills/<dir>/SKILL.md), lockfile claude-lock.json (commit it); CI: ant apply --yes . with WIF.
ant beta:sessions connect --session-id … [--web] [--no-browser] [-v] — live transcript, Enter sends user.message, Esc interrupts, tool-approval prompts.

# Not covered here

Claude Code (claude CLI) is a different tool (agentic coding); Agent SDK and Managed Agents are documented elsewhere in this atlas.

# Anthropic ant CLI