> ## Documentation Index
> Fetch the complete documentation index at: https://docs.autonomyai.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Autonomy MCP

> Drive the Autonomy coding platform from any MCP client.

The **Autonomy MCP server** exposes the Autonomy coding platform - projects, coding tasks, setup, and Full Mode - as [Model Context Protocol](https://modelcontextprotocol.io) tools that any MCP host can call: Claude Code, Claude Desktop, and others.

Point your AI agent at it and it can pick a project, kick off a coding task, watch it run, pull the diff, and send the work to your developers - all without leaving the editor.

<CardGroup cols={2}>
  <Card title="Quickstart" icon="rocket" href="/getting-started/quickstart">
    Connect the MCP to your client in two minutes.
  </Card>

  <Card title="Tool Reference" icon="wrench" href="/tools/overview">
    Every tool, grouped by domain, with params and returns.
  </Card>

  <Card title="Run a task" icon="play" href="/concepts/tasks">
    How a coding task flows from instruction to shipped PR.
  </Card>

  <Card title="Full Mode" icon="layer-group" href="/concepts/full-mode">
    Run your whole app against your staging backend.
  </Card>
</CardGroup>

## What you can do

* **Manage projects** - list connected repos, initialize a project's workspace, add new ones.
* **Run coding tasks** - start a task from a natural-language instruction, poll it, reply to it, read what the agent did, and pull the diff.
* **Set a project up** - configure install / validate / dev commands and environment variables so the agent can build in it.
* **Enable Full Mode** - serve the project's whole front end in Autonomy while proxying API calls to your existing staging backend.
* **Ship** - finalize a task and open or update the pull request for your developers.

## How it works

The server is a **thin adapter** over Autonomy's backend (FEI): every tool maps to one backend endpoint. No agent logic lives in the MCP server - orchestration, sandboxes, and PR creation all happen in the platform. The server authenticates the caller and forwards the request.

```
MCP host ──MCP──► Autonomy MCP ──HTTP + OAuth──► Autonomy platform ──► sandboxes / PRs
```

Because it is stateless, the conversation lives in your client and all durable state lives in the platform - so tasks you start over MCP show up in [Studio](https://studio.autonomyai.io) and vice versa.

<Note>
  New to Autonomy? Start with the [Quickstart](/getting-started/quickstart), then read [Run a coding task](/concepts/tasks).
</Note>