We’ve added a CLI for Claude Platform to make every API endpoint runnable from your terminal. Call the Messages API, stand up Claude Managed Agents, pipe results straight into your shell. The ant CLI is well understood by coding agents (Claude Code) using the claude-api skill.

For interactive login, the CLI supports "ant auth login". This runs a browser OAuth flow, scopes the token to a workspace, and stores it for you. That one credential is then used by the CLI (and even the SDKs) to authenticate your API calls.

Every API resource is a subcommand, including messages, models, and files. Build requests from typed flags or piped YAML, inline a file with @.path, reshape any response with --transform. Output as json, yaml, jsonl, or explore (default) for a TUI you can fold and search.

Use the ant CLI to create and update Claude Managed Agents, locally or from CI. Version control your agent as YAML in Git and have CI run "ant beta:agents update" to sync changes to the Claude Platform.

The CLI can kick off a Managed Agents session, send it an event, and the agent goes to work: reasoning, calling tools, working toward an answer. When it's done (or stuck), pull the trace: every event, every tool call, every decision is retrievable from the same CLI.

Claude Code knows how to use ant using the built-in "/claude-api" skill Ask it to list your sessions, upload a folder of PDFs, or debug a run. It shells out to the CLI and reads the results back without any glue code. Install with brew, curl, or go:

Running agents straight from the terminal is insane. about to abuse this immediately!!

Omg can you slow it down

Is there anything claude can’t do??? Wtf

It's more than a feature it's the game changer for everyone.

Step 1: Give AI a CLI. Step 2: Ask it to fix your code. Step 3: Pretend you did it yourself. 😎

ant. they named it ant. somewhere a pm is fighting for their life in a thread about why claude-cli was taken or why this won't confuse every search for apache ant build tool from 2003

Ant CLI just dropped and suddenly Claude devs have a better terminal experience than most of us 😂 Piping agents straight into shell + native Claude Code integration = dangerous productivity. Well played @ClaudeDevs

This is exactly what the Claude API workflow was missing. Being able to pipe results straight into shell scripts makes agentic automation so much cleaner. The claude-api skill integration with Claude Code is the cherry on top. More CLI tools like this please.

The “open docs, copy curl, paste, pray” workflow might finally be getting replaced.

Wait!! It means every ai product we are trying to create can be one prompted with Claude

ANThropic 🐜

Antropians 🐜

Will it cost me $1000 to use or can I use my subscription to use this?

The terminal quietly became the new IDE. Piping agent output straight into your shell beats bolting on another dashboard. The teams who win with AI won't have the prettiest UI — they'll have made it composable. Big step.

The real question - is this covered by the subscription or not?

the pipe results into shell is going to be underrated for a while. think: git diff | ant messages --prompt 'review this' | tee review.txt. makes claude a first-class citizen in your build pipeline, not just a chat window you tab over to.

version control your agent as YAML. CI syncs it automatically. pipe results straight into your shell. the API just became a first-class terminal citizen. about time.

A CLI is huge for people learning by doing. Even for non-coders, terminal examples make the workflow visible: run one request, see the response, change one thing, repeat. That is much less mysterious than trying to understand APIs in theory.

Ohh thou anthropic masters, and what shall our allowance be for this? 🥹 *cough* cocksuckers *cough*

@grok Can you explain plainly why this is a benefit?

piping agent outputs straight into the shell is where things start feeling less like “tools” and more like runtime composition. CLI becoming an agent interface is a pretty big shift

The ant CLI meets Standard networking laws dictate I have to integrate this now. 🐜 I’m 10 days into building an analytics and AI-visibility platform, and being able to trigger managed agents and pipe results via terminal is going to make automation so much cleaner. Sick update!

CLI access is where APIs start feeling like infrastructure, not demos. Add two guardrails early: log every request shape, and keep one replayable curl for each endpoint. Agent failures get much less mysterious. Follow us for practical API workflow notes.

piping agent outputs straight into the shell is where things start feeling less like “tools” and more like runtime composition. CLI becoming an agent interface is a pretty big shift....

@ClaudeDevs nice! this'll make testing so much smoother. was a pain switching between tools before. excited to see how it speeds up my workflow.

At this rate my job description will become "approve what the agents already did." 😂

Just build it without the managed agents and no api charges

ANThropic dropping the ultimate terminal power tool. Next up: ant deploy-my-entire-startup Seriously though, the GitOps flow for Managed Agents is 🔥

@FounderModes Very useful

Piping Managed Agents directly into the shell is an absolute game-changer for terminal-centric workflows. Skipping the boilerplate code just to test endpoints or orchestrate quick multi-agent loops saves a massive amount of friction during early-stage prototyping. The ant CLI is going to speed up developer velocity significantly. 💻🔥

Like the name, we also have product named

Ant?? Welcome back

piping managed agents directly into the shell is exactly the workflow we needed. Thank you @claudeai team.

The claude-api skill is the real unlock here — agents can now self-serve the whole platform without you hand-writing wrappers. CLI becomes the interface layer your agent reads, not just the one you type into.