AI/TLDR

GitHub MCP Server

GitHub's own MCP server: repos, issues, PRs, Actions and security alerts as agent tools

Workflow AutomationOpen source
Language
Go
License
MIT

Overview

The GitHub MCP Server is GitHub's official Model Context Protocol server. It hands an AI agent, assistant or chatbot a set of GitHub tools: reading repositories and code files, creating and updating issues and pull requests, monitoring Actions workflow runs and analysing build failures, reading code-scanning, Dependabot and secret-scanning alerts, and working with discussions, notifications, projects, gists and organisation data.

There are two ways to run it. The remote server is hosted by GitHub at `https://api.githubcopilot.com/mcp/` and is the quickest path for any host that supports remote MCP with OAuth — VS Code 1.101+, Claude Desktop, Cursor, Windsurf and others — with one-click install buttons in the README. The local server is a Go binary, published as a Docker image at `ghcr.io/github/github-mcp-server`, authenticated with OAuth or a `GITHUB_PERSONAL_ACCESS_TOKEN`; installation guides cover Copilot CLI, Claude, Codex, Cursor, OpenCode, Windsurf, JetBrains, Visual Studio, Eclipse and Xcode.

Tool surface is controlled rather than all-or-nothing. Functionality is grouped into toolsets — `context`, `repos`, `issues`, `pull_requests`, `actions`, `code_security`, `secret_protection`, `dependabot`, `discussions`, `gists`, `git`, `governance`, `labels`, `notifications`, `orgs`, `projects`, `security_advisories`, `stargazers`, `users` and more — enabled with `--toolsets` or the `GITHUB_TOOLSETS` environment variable, and individual tools can be allow-listed with `--tools`. Enabling only what you need helps tool choice and keeps the context small. A `--read-only` flag drops every write tool and takes priority over explicit `--tools` entries. The project is MIT licensed.

What it does

  • Official, GitHub-maintained MCP server, available both hosted (remote) and self-run (local)
  • Toolsets for repositories, issues, pull requests, Actions, code security, secret protection, Dependabot, discussions, projects and more
  • Fine-grained control with `--toolsets` / `GITHUB_TOOLSETS` and `--tools` / `GITHUB_TOOLS`
  • `--read-only` mode that removes every write tool, overriding explicit tool requests
  • OAuth or personal-access-token authentication, so the agent only gets the scopes you grant
  • Distributed as a Go binary and a public Docker image, with install guides for a dozen MCP hosts

Getting started

Start with the remote server if your host supports remote MCP; otherwise run the Docker image locally. Then narrow the tool surface.

Use the remote server

Point your MCP host at GitHub's hosted endpoint — VS Code 1.101 or later handles the OAuth flow for you.

jsonjson
{
  "servers": {
    "github": {
      "type": "http",
      "url": "https://api.githubcopilot.com/mcp/"
    }
  }
}

Or run it locally with Docker

The image is public. This example registers the local server with Claude Code using a personal access token.

bashbash
claude mcp add github -e GITHUB_PERSONAL_ACCESS_TOKEN=$GITHUB_PAT -- \
  docker run -i --rm -e GITHUB_PERSONAL_ACCESS_TOKEN ghcr.io/github/github-mcp-server

Enable only the toolsets you need

Fewer tools means better tool choice and less context spent on schemas.

bashbash
github-mcp-server --toolsets repos,issues,pull_requests,actions,code_security

# or via the environment
GITHUB_TOOLSETS="repos,issues,pull_requests" ./github-mcp-server

Lock it to read-only

Write tools are skipped entirely, even if named in `--tools`.

bashbash
./github-mcp-server --read-only

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

When to use it

  • Let a coding agent open, review and update pull requests in the repositories you grant it
  • Triage issues and project boards from an assistant instead of the web UI
  • Have an agent read a failing Actions run and explain or fix the build
  • Give a read-only agent access to code scanning, Dependabot and secret-scanning findings for security review

How GitHub MCP Server compares

GitHub MCP Server alongside other open-source workflow automation tools AI/TLDR tracks, ranked by GitHub stars.

ToolStarsWhat it does
n8n★ 205kA self-hostable workflow automation tool with a visual node editor that connects 400+ apps and APIs and adds native AI steps, letting technical teams build automations without writing most of the glue code.
Huginn★ 50kA self-hosted system of "agents" that watch websites, feeds, and events and take automated actions on your behalf, similar to a private IFTTT.
GitHub MCP Server★ 33kGitHub's own MCP server: repos, issues, PRs, Actions and security alerts as agent tools
gws (Google Workspace CLI)★ 31kA command-line tool for every Google Workspace API, built at runtime from Google’s Discovery Service, returning structured JSON and shipping 40+ agent skills. Not an officially supported Google product.
Kestra★ 28.1kAn event-driven orchestration platform that defines data, AI, and infrastructure pipelines as declarative YAML and runs them through a web UI.
Activepieces★ 24.5kAn open-source Zapier alternative where you build automations from reusable TypeScript "pieces", with support for AI agents and self-hosting.
n8n-MCP★ 22.9kAn MCP server that gives coding assistants structured access to n8n's node catalogue, schemas, docs and workflow templates so they can build and validate n8n workflows.
Trigger.dev★ 16.3kAn open-source TypeScript platform for durable background tasks and AI agents — no timeouts, with retries, queues, waitpoints for human approval, realtime streaming and per-run tracing.