github-copilot-cli-mcp-server
Wraps GitHub Copilot CLI as an MCP server, allowing AI agents to run Copilot conversations, resume sessions, list sessions, and respond to permission prompts for code generation, modification, and analysis tasks.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@github-copilot-cli-mcp-server Fix the bug in src/main.ts"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
github-copilot-cli-mcp-server
⚠️ ALPHA VERSION — This project is experimental and in early development. Testing across different environments is limited. Use at your own risk and expect breaking changes.
A Node.js/TypeScript project that wraps GitHub Copilot CLI as a Model Context Protocol (MCP) server.
Use Copilot CLI as a tool from any MCP-compatible client (VSCode, OpenClaw, Claude Desktop, etc.).
Features
Complete Copilot conversations in a single MCP call: Send a prompt → receive results in one tool call
Session resumption: Continue previous conversations using session IDs
Permission mode selection: Interactive (user confirmation) / Autonomous (auto-approve)
Model selection: Use any model supported by Copilot
Working directory specification: Set cwd for tasks that require file access
Related MCP server: mcp-github-copilot
Prerequisites
Node.js 20.0.0 or higher
GitHub Copilot CLI installed and authenticated
npm install -g @github/copilot-cliGitHub Copilot subscription (Individual, Business, or Enterprise)
Installation
From source
git clone https://github.com/slcwahn/github-copilot-cli-mcp-server.git
cd github-copilot-cli-mcp-server
npm install
npm run buildQuick start
npm run devMCP Tools
run_copilot_conversation
Runs a Copilot CLI conversation with a prompt.
Parameters:
Name | Type | Required | Description |
| string | ✅ | Prompt to send to Copilot |
| string | AI model (e.g., | |
| string | Working directory | |
| string[] | List of tools to allow | |
| string[] | Additional directories to grant access | |
| number | Timeout (default: 300000ms = 5 minutes) | |
| string | Permission mode: |
Example:
{
"name": "run_copilot_conversation",
"arguments": {
"prompt": "Fix the bug in src/main.ts",
"cwd": "/path/to/project",
"model": "claude-sonnet-4"
}
}resume_copilot_session
Resumes a previous session to continue the conversation.
Parameters:
Name | Type | Required | Description |
| string | ✅ | Session ID (UUID) |
| string | ✅ | Follow-up prompt |
| string | AI model | |
| string | Working directory | |
| number | Timeout |
Example:
{
"name": "resume_copilot_session",
"arguments": {
"session_id": "abc12345-1234-5678-9abc-def012345678",
"prompt": "Now add tests for those changes"
}
}list_copilot_sessions
Lists resumable Copilot CLI sessions.
respond_to_copilot
Responds to Copilot's permission prompts in Interactive mode.
Parameters:
Name | Type | Required | Description |
| string | ✅ | Pending session ID |
| string | ✅ | Response ( |
Configuration
VSCode
Add the following to your .vscode/mcp.json file:
{
"servers": {
"github-copilot-cli": {
"type": "stdio",
"command": "node",
"args": ["/path/to/github-copilot-cli-mcp-server/dist/index.js"],
"env": {
"COPILOT_PERMISSION_MODE": "interactive"
}
}
}
}Or use the Command Palette: MCP: Add Server → stdio → enter the configuration above.
Tip: Use
${workspaceFolder}to specify relative paths based on your workspace.
For global configuration, use the Command Palette: MCP: Open User Configuration to add it to your user profile.
OpenClaw (mcporter)
OpenClaw supports MCP servers through the mcporter skill.
Register via mcporter CLI
# Register the server
mcporter config add github-copilot-cli \
--command node \
--arg /path/to/github-copilot-cli-mcp-server/dist/index.js \
--env COPILOT_PERMISSION_MODE=autonomous
# Verify registration
mcporter list
# Check tool schema
mcporter list github-copilot-cli --schema
# Call a tool directly
mcporter call github-copilot-cli.run_copilot_conversation prompt="Fix the bug in main.ts"Edit mcporter config file directly
~/.mcporter/mcporter.json or project-level config/mcporter.json:
{
"servers": {
"github-copilot-cli": {
"transport": "stdio",
"command": "node",
"args": ["/path/to/github-copilot-cli-mcp-server/dist/index.js"],
"env": {
"COPILOT_PERMISSION_MODE": "autonomous"
}
}
}
}Claude Desktop
Add to claude_desktop_config.json:
{
"mcpServers": {
"github-copilot-cli": {
"command": "node",
"args": ["/path/to/github-copilot-cli-mcp-server/dist/index.js"]
}
}
}Environment Variables
Variable | Default | Description |
| (auto-detect) | Path to Copilot CLI binary |
|
| Permission mode: |
Permission Handling
Copilot CLI may request user approval for file modifications, shell command execution, and other actions. This MCP server supports two permission modes:
Autonomous Mode (default)
COPILOT_PERMISSION_MODE=autonomousRuns Copilot CLI with
--allow-all-tools --no-ask-userflagsAuto-approves all permissions and completes without user prompts
Best for: Trusted tasks, automation pipelines, CI/CD
Interactive Mode
COPILOT_PERMISSION_MODE=interactiveRuns Copilot CLI with PTY for interactive I/O
Returns
needsInput: truein the MCP response when a permission prompt is detectedThe MCP client (user or agent) responds via the
respond_to_copilottool
Interactive Mode Flow:
MCP Client MCP Server Copilot CLI
│ │ │
│ run_copilot_conversation │ │
├─────────────────────────────►│ spawn (PTY) │
│ ├───────────────────────────►│
│ │ │
│ │ "Modify this file?" │
│ │◄───────────────────────────┤
│ { needsInput: true, │ │
│ question: "Modify..." } │ │
│◄─────────────────────────────┤ │
│ │ │
│ respond_to_copilot("yes") │ │
├─────────────────────────────►│ write "yes\n" │
│ ├───────────────────────────►│
│ │ │
│ │ (complete) │
│ │◄───────────────────────────┤
│ { output: "..." } │ │
│◄─────────────────────────────┤ │Note: Interactive mode requires
node-pty(optional dependency). If not installed, it automatically falls back to autonomous mode.
Architecture
MCP Client (VSCode / OpenClaw / Claude Desktop)
│
│ stdio (JSON-RPC)
▼
┌──────────────────────────────────────────────┐
│ github-copilot-cli-mcp-server │
│ │
│ ┌────────────────────────────────────────┐ │
│ │ MCP Server (stdio) │ │
│ │ - run_copilot_conversation │ │
│ │ - resume_copilot_session │ │
│ │ - list_copilot_sessions │ │
│ │ - respond_to_copilot │ │
│ └──────────────┬─────────────────────────┘ │
│ │ │
│ ┌──────────────▼─────────────────────────┐ │
│ │ Permission Handler │ │
│ │ (autonomous / interactive) │ │
│ └──────────────┬─────────────────────────┘ │
│ │ │
│ ┌──────────────▼─────────────────────────┐ │
│ │ Copilot Runner │ │
│ │ (spawn / PTY) │ │
│ └──────────────┬─────────────────────────┘ │
│ │ │
│ ┌──────────────▼─────────────────────────┐ │
│ │ Session Manager │ │
│ │ (session metadata + pending input) │ │
│ └────────────────────────────────────────┘ │
└──────────────────────────────────────────────┘
│
│ spawn / PTY
▼
copilot -p "prompt" -s [--allow-all-tools | interactive]Development
# Development mode
npm run dev
# Build
npm run build
# Type check
npm run typecheck
# Test
npm testHow It Works
The MCP client calls the
run_copilot_conversationtoolDepending on the permission mode:
Autonomous: Runs
copilot -p "<prompt>" -s --allow-all-tools --no-ask-userInteractive: Runs with PTY, forwarding permission prompts to the MCP client
Copilot CLI performs the task (code generation, modification, analysis, etc.)
Returns the output as an MCP response upon completion
If a session ID is available, the session can be resumed via
resume_copilot_session
Copilot CLI Options Used
Flag | Purpose |
| Run prompt in non-interactive mode |
| Silent mode (response only, no stats) |
| Auto-approve all tools (autonomous mode) |
| Autonomous operation without prompts (autonomous mode) |
| Ignore AGENTS.md and similar files |
| Disable ANSI colors |
| Disable terminal alternate screen |
| Resume a session |
| Select a model |
| Grant access to additional directories |
License
MIT
References
Available Tools
4 toolslist_copilot_sessionsA
List available Copilot CLI sessions that can be resumed. Also shows any sessions waiting for permission input.
Inputs: no named parameters.
respond_to_copilotA
Respond to a Copilot CLI permission question in interactive mode. When Copilot asks for permission (e.g., 'Allow file modification?'), use this tool to send the user's response (e.g., 'yes', 'no', or custom text).
Inputs: response (required), session_id (required).
resume_copilot_sessionA
Resume a previous Copilot CLI session by session ID. This continues an existing conversation with additional context or follow-up questions.
Inputs: prompt (required), session_id (required), cwd, model, +2 more.
run_copilot_conversationA
Execute a prompt with GitHub Copilot CLI. Runs Copilot in non-interactive mode (-p) and returns the complete response. Use this for one-shot tasks like code generation, explanation, debugging, etc. In interactive permission mode, may return with needsInput=true if Copilot asks a permission question — use respond_to_copilot to answer.
Inputs: prompt (required), cwd, model, add_dirs, +3 more.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 4 tools
Each tool has a clearly distinct role: run starts a new one-shot conversation, list shows available sessions, resume continues an existing session by ID, and respond answers permission prompts. No two tools overlap in purpose, and the descriptions reinforce the boundaries.
All tool names follow a consistent verb_noun pattern: resume_copilot_session, run_copilot_conversation, list_copilot_sessions, respond_to_copilot. The pattern makes the action and target of each tool immediately predictable.
Four tools is well-scoped for a thin MCP server wrapping GitHub Copilot CLI. Each tool covers a necessary part of the session workflow—start, list, resume, and respond—without redundant or filler tools.
The core session lifecycle is covered: starting, listing, resuming, and handling permission input. Minor gaps exist, such as no explicit way to terminate a session or fetch full session transcripts, but these are not critical for the server's apparent purpose.
Maintenance
Related MCP Connectors
Shared memory MCP server for Claude Code, Codex, Gemini CLI and Cursor. Cited, editable.
GitHub Copilot MCP: code, repo, PR, workflow tools for AI assistants.
Remote MCP server for supportsheep: run AI interviews and manage support content for your blog.
Remote MCP server for supportsheep: run AI interviews and manage support content for your blog.
Related MCP Servers
- AlicenseCqualityDmaintenanceConnects MCP-enabled editors to GitHub Copilot CLI for non-interactive code analysis, batch processing, and code review.1028 npm5MIT
- AlicenseAqualityCmaintenanceMCP server for GitHub Copilot that allows querying any Copilot model programmatically using existing Copilot CLI credentials, with support for file attachments and model discovery.226 npmMIT
- AlicenseAqualityBmaintenanceMCP server for opencode that queries GitHub Copilot models via a persistent opencode server, requiring no API keys.288 npm3MIT
- AlicenseBqualityDmaintenanceWraps the copilot CLI into MCP tools, allowing third-party MCP clients to invoke Copilot through prompt and command tools.2GPL 3.0