Anthropic Messages API and Claude Managed Agents.

Optional. Prefer workspace secrets ANTHROPIC_API_KEY and GITHUB_TOKEN; you can also pass keys here.

configure

apiKey

GitHub token used when run mounts a repo (prefer workspace secret GITHUB_TOKEN).

Send a Messages API request and return the first text block.

The model that will complete your prompt.

An array of message objects representing the conversation.

The maximum number of tokens to generate.

An object describing metadata about the request.

Sequences that will cause the model to stop generating completion text.

An array of system message objects, including cache control.

Amount of randomness injected into the response.

How the model should use the provided tools.

Definitions of tools that the model may use.

Only sample from the top K options for each subsequent token.

Create (or reuse) a Managed Agents agent and environment, run a session with the given prompt, wait until idle, and return a summary.

Name for a newly created agent (and environment).

Tools array for a newly created agent. Defaults to the full agent toolset.

Reuse an existing agent id instead of creating one.

Reuse an existing environment id instead of creating one.

Max wait for the session to become idle, in milliseconds (default 15 minutes).

GitHub repo to mount (https://github.com/owner/name or owner/name).

GitHub token for clone/push (required when repo is set).

Raw Managed Agents session resources array (advanced).

Optional vault id list for MCP credentials.