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

# Drive Fermata from Claude Code over MCP

> Fermata runs a local MCP server, so a Claude Code session in your terminal can read your pieces and move them along without you opening the app.

Fermata already runs your own Claude, so the MCP servers you configured work inside Fermata sessions with no setup. This page is about the other direction. Fermata itself is an MCP server, the control plane, and any MCP client can connect to it. A Claude Code session in your terminal is the usual one.

Once it is connected, you talk to Claude and Claude drives Fermata. It lists your projects, pieces, and sessions. It reads a piece's spec, strategy, summary, learnings, analysis, work review, code review, and agent handoffs. It files drafts into your Backlog, from a spec you wrote or from a Linear or GitHub issue key, and releases them. It walks a piece through its flow: answering the interview, approving the spec and the strategy, starting and stopping agents, asking for changes, opening the pull request, and marking it done.

## Why you would use it

**You stay where you are.** You are already in a terminal with Claude. "What needs me in Fermata?" is one question there, not a window switch and a scan of the board.

**Bulk work is one sentence.** The app moves one card at a time. Claude can loop. "Start these five bugs as Auto pieces" files five drafts and releases them; "open a PR for every finished piece" walks the list and opens each one. See [what you can do](https://docs.fermata.run/mcp/what-you-can-do.md) for the requests that map cleanly onto the tools.

**Another agent can check on its own work.** A Claude session that filed a piece can come back later, read its phase, and pick up the pull request link, without you relaying anything.

## Local only

The server listens on `127.0.0.1`, the loopback address, so nothing on your network can reach it. Every request carries a token in the URL, and a request with the wrong token is refused before it does anything. Nothing goes through the cloud: the client talks straight to the app on your Mac, and the server stops when Fermata quits.

## It does not go around you

The control plane is another way to press the buttons you already have, not a way past them. A piece on the Manual lane still stops at every phase. A build agent on the Safe profile still raises its tool approvals inside Fermata. Stopping agents and marking a piece done always ask you on your Mac first, and you can make runs and pull requests ask too. Merging is not on the list at all: Fermata opens the pull request, and you merge. [Consent and safety](https://docs.fermata.run/mcp/consent-and-safety.md) covers each rule.

## Next

> [card] [Connect a client](https://docs.fermata.run/mcp/connect.md)
>
> Turn the server on, register it with Claude Code, and check it answers.

> [card] [What you can do](https://docs.fermata.run/mcp/what-you-can-do.md)
>
> Requests you type to Claude, and the tools they become.

> [card] [Consent and safety](https://docs.fermata.run/mcp/consent-and-safety.md)
>
> What runs on the call, what waits for a card, and what is never exposed.

> [card] [Tool reference](https://docs.fermata.run/reference/mcp-tools.md)
>
> Every tool and every piece action, with its arguments.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.