GitHub MCP Server
This official MCP exposes the GitHub API to an agent: creating and commenting on issues, managing pull requests, checking GitHub Actions workflows, without going through the `gh` CLI or the web interface for every operation. It exists as both a local (Docker) version and a hosted remote version for GitHub Copilot.
Key advantages
- Officially maintained by GitHub
- Broad coverage of the GitHub API
- Complements the `gh` CLI for conversational use
Who it's for, and why
Development teams who want their AI agent to interact directly with their GitHub repositories.
- Managing issues and pull requests from the agent
- Checking the status of GitHub Actions workflows
- Automating repository management tasks
Getting started
- 1Locally via Docker: `claude mcp add github -e GITHUB_PERSONAL_ACCESS_TOKEN=<token> -- docker run -i --rm -e GITHUB_PERSONAL_ACCESS_TOKEN ghcr.io/github/github-mcp-server`.
- 2Alternative without Docker: build the binary yourself with `go build` in `cmd/github-mcp-server`.
- 3Remote option: the hosted endpoint `https://api.githubcopilot.com/mcp/`, restricted to clients supporting GitHub Copilot OAuth (VS Code 1.101+).
Things to watch out for
- The local version requires Docker (or a Go build): one extra technical step compared to purely remote MCPs like Vercel or Supabase.
- A poorly scoped personal access token (PAT) gives the agent the same rights as the associated GitHub account: limiting the PAT's scopes is recommended.
Frequently asked questions
Do I need Docker to use this MCP?
For the documented local install, yes. An alternative exists by building the binary yourself (`go build`), or by using the remote endpoint restricted to GitHub Copilot.
Can this MCP close pull requests automatically without confirmation?
It exposes the actions allowed by the provided token; it's up to the user to configure a token with limited scopes and to keep human oversight on sensitive actions (merging, closing a PR).
Does it work with Claude Code?
Yes, via `claude mcp add` with the Docker command provided in the project's official README.
Similar tools
Worth exploring too
Claude Code
Anthropic's command-line agent, able to read, write and execute code directly inside a real project.
Vercel MCP
Vercel's MCP server: check deployments, logs and projects directly from an AI agent.
Playwright
Microsoft's browser testing and automation framework, used here to capture real screenshots.