MCP Integration

Overview

rft includes a built-in MCP (Model Context Protocol) server. This allows AI agents like Claude Code to manage Docker Compose worktree stacks directly.

Setup

Add to your Claude Code MCP configuration (.mcp.json or project settings):

{
  "mcpServers": {
    "rft": {
      "command": "rft",
      "args": ["mcp"]
    }
  }
}

Available Tools

The MCP server exposes 6 tools:

Tool	Description
`rft_list`	List all worktrees with ports and status
`rft_start`	Start Docker Compose stacks
`rft_stop`	Stop running stacks
`rft_restart`	Restart stacks
`rft_promote`	Transfer changes from a worktree
`rft_clean`	Full cleanup of all resources

Starting the Server

The MCP server runs via stdio transport:

rft mcp

This is intended to be called by the AI agent, not run manually. The agent communicates with the server over stdin/stdout using the MCP protocol.

Use Cases

With MCP integration, an AI agent can:

Check worktree status before making changes
Start and stop stacks needed for testing
Promote changes from a feature branch worktree
Clean up after completing a task