Let your agent edit with the MCP server

The Coedit MCP server lets your own AI agent read a project and its comments and propose changes, without leaving your tool. It exposes the same actions as the CLI as tools the agent can call. Because your agent runs on your own model, this route costs you nothing extra in Coedit.

What is the Coedit MCP binary?

It is a small server that ships inside the Coedit repository at mcp/bin/server.mjs. It speaks the Model Context Protocol (an open standard for connecting AI tools to data) over your machine, and calls the same Coedit REST API the app and CLI use. "Adding the server" just means telling your AI tool to run that one file with Node. There is nothing to deploy.

Before you start

• Node.js 18 or newer installed (check with node --version).
• The Coedit repository on your machine (you need the mcp/ folder). If you do not have it, clone it from your Coedit source.
• A running Coedit server to point at: the local dev server at http://localhost:3000, or your hosted Coedit URL (for example https://app.coeditapp.com).

Step 1: Install the server

Install the server's one dependency, then note its absolute path (you will paste it into your AI tool).

cd /path/to/Coedit/mcp
npm install

# Print the absolute path to the server file, you'll need it next:
node -e "console.log(require('path').resolve('bin/server.mjs'))"
# e.g. /Users/you/Coedit/mcp/bin/server.mjs

Step 2: Add it to your AI tool

The idea is the same in every client: run node <path-to-server.mjs> with the COEDIT_API environment variable set. Pick your tool below.

Claude Code (terminal). Use the built-in claude mcp add command. Replace the path with the one you printed in Step 1:

claude mcp add --env COEDIT_API=http://localhost:3000 \
  --transport stdio coedit \
  -- node /absolute/path/to/Coedit/mcp/bin/server.mjs

Claude Desktop, Cursor, or any JSON-configured client. Add a coedit entry to the client's MCP config file (Claude Desktop: claude_desktop_config.json; Cursor: .cursor/mcp.json), then restart the app:

{
  "mcpServers": {
    "coedit": {
      "command": "node",
      "args": ["/absolute/path/to/Coedit/mcp/bin/server.mjs"],
      "env": { "COEDIT_API": "http://localhost:3000" }
    }
  }
}

Step 3: Confirm it's connected

1. 1Restart your AI tool (or, in Claude Code, run /mcp to see server status).
2. 2Check that the coedit tools appear (for example coedit_list_projects).
3. 3Ask your agent to list your Coedit projects. If it returns them, you are connected.