🤖 GitHub Copilot MCP Server

Welcome to the Model Context Protocol (MCP) server for the GitHub Copilot Agentic CLI.

What does this actually mean? It means you can expose Copilot's shiny new autonomous capabilities as tools to other AI agents. That's right—we're making AI agents talk to AI agents. It's subagents all the way down. 🐢🐢🐢

Use this to delegate complex coding tasks, review code, or just ask questions to Copilot while your main agent (like Claude, Cursor, or Codex) sits back, sips a digital margarita, and manages the big picture.

🧠 Why is this cool? (Use Cases)

Ever wanted to tell an AI to tell another AI what to do?

• The Ultimate Subagent Workflow: Configure this MCP server in your main AI agent (like Claude Desktop or Cursor). Then, when you ask your main agent to build a feature, it can say, "Hey Copilot, you go write the CSS for this," while the main agent handles the backend logic.
• Save the Big Brain Tokens: Got a simple refactor, docstring generation, or boilerplate task? Delegate it to Copilot!
• Leverage the Free Tier: You can totally use the Copilot Free Tier to handle the grunt work! Let Copilot churn out basic file scaffolding as a subagent while you use your premium Claude tokens for the heavy architectural lifting. It's free real estate! 🏠
• Parallel Processing: Your main agent can fire off a background copilot_delegate_task, continue working on something else, and check the results when Copilot finishes.

🛠️ Prerequisites

• Node.js (v18+, you know the drill)
• The modern GitHub Copilot CLI (copilot binary) installed and authenticated on your machine.

🧰 Tools Provided

Here's the arsenal you get out of the box:

copilot_delegate_task

Delegates an autonomous task to the Copilot CLI. It runs in the background and returns the result once it's done. Copilot runs with full permissions (--yolo) and without asking for user confirmation (--no-ask-user), allowing it to just get the job done autonomously. Parameters:

• prompt (string, required): The detailed task instruction or prompt. Tell it what to do!
• cwd (string, optional): Absolute path to the working directory. Defaults to wherever you are right now.
• model (string, optional): The AI model to use (e.g., gpt-4o, claude-3.5-sonnet).

copilot_ask

Ask Copilot a question or request an explanation without giving it permission to touch your files. Perfect for code explanation, debugging, or architectural brainstorming. Parameters:

• prompt (string, required): The burning question you have.
• cwd (string, optional): Absolute path to the working directory for context.
• model (string, optional): The AI model to use.

copilot_resume_session

Resume a previous Copilot session or task using its session ID. Because sometimes one prompt just isn't enough to get it right. Parameters:

• prompt (string, required): The follow-up instructions.
• sessionId (string, optional): The UUID of the session to resume. If omitted, it just grabs the most recent one.
• cwd (string, optional): Absolute path to the working directory.
• model (string, optional): The AI model to use.

copilot_review_code

Unleash Copilot to relentlessly judge your code. It will autonomously analyze a file or directory and return a detailed report without changing anything. Parameters:

• target (string, required): The file or directory path to review (e.g., src/index.ts or .).
• focusArea (string, optional): Specific thing to nitpick (e.g., 'security', 'performance', 'readability', 'bugs').
• cwd (string, optional): Absolute path to the working directory.

copilot_list_models

Need to know what brains are available? This returns a list of commonly supported AI models by the GitHub Copilot CLI.

🏗️ Architecture & Project Structure

The project has been built to be scalable, readable, and highly maintainable:

• src/index.ts: The main entry point linking the MCP transport layer.
• src/server.ts: Configures the MCP server instance and registers our glorious tools.
• src/tools/*: Each tool is isolated in its own file (e.g., ask.ts, delegateTask.ts) to ensure single-responsibility.
• src/utils/exec.ts: A centralized abstraction for reliably wrangling the copilot CLI process.

🏃‍♂️ Running Locally

To test or run locally without installing globally:

npm install
npm run build
node build/index.js

🔌 Running via npx (for Cursor, Claude Desktop, etc.)

You can effortlessly plug this server into any standard MCP client using npx. Just add this to your MCP config:

{
  "mcpServers": {
    "copilot-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "@devwithadi/copilot-mcp@latest"
      ]
    }
  }
}

Pro-tip: If you are hacking on this locally, use the local path instead:

{
  "mcpServers": {
    "copilot-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "/absolute/path/to/copilot_mcp"
      ]
    }
  }
}

Enjoy delegating your delegations! 🎉