Let your agent set up AgentTrunk

The documentation site’s MCP server answers questions about these public guides. It does not sign in to AgentTrunk, upload context, or read private workspaces. Use the CLI or SDK below for authenticated product operations.

Give a terminal-capable agent this task:

Set up AgentTrunk for this project.
Read https://docs.agenttrunk.ai/quickstart.md and
https://docs.agenttrunk.ai/authentication.md first.
Reuse existing authorized credentials if available. Otherwise initiate
human-approved registration and give me the verification link.
Do not read my inbox, cookies, or passwords. Ask me for the approval code.
List workspaces and ask which one to use if my intent is ambiguous.
Do not upload context or change production as part of this setup task.
Report installation, authentication, and workspace access separately.

If your human is new

They sign up and complete organization onboarding. Existing users sign in. Then the agent completes the approval flow described in authentication.

Discover only what you need

Use /docs/llms.txt to choose a page, then append .md to read that guide as Markdown. Use the API reference for exact fields rather than guessing routes. The public source can be inspected without signing in.

Skills and MCP

The portable skill teaches the workflow; it is not a credential. Install it using the integration guide.

There is no hosted AgentTrunk service MCP endpoint currently documented. A documentation search tool is not an upload or release API. Use the CLI, SDK, or REST API for those actions; do not invent MCP signup or connection URLs.