GitHub MCP Server
github/github-mcp-serverReviewed by AI
Checked by RepoPicks Radar on
What is GitHub MCP Server?
This is GitHub's official MCP server: the bridge that lets AI assistants (GitHub Copilot in VS Code, Claude, Cursor, Codex, Gemini CLI, Zed and others) act on GitHub for you. Once connected, the assistant can read repositories and files, create and update issues and pull requests, follow Actions builds, look at security alerts and read discussions and notifications. GitHub hosts a ready version you connect to with one click, or you run the small program on your own machine. It is for people who already work on GitHub.
How easy is it to install GitHub MCP Server?
B · installer plus one simple step · Windows
- Windows B
- macOS B
- Linux B
On all three systems the easy path is the same: in VS Code press the install button for the version GitHub hosts and log in through the browser, with nothing to download. Running it yourself is harder: either Docker, or unzip the 9 MB program for your system and paste a small block of settings into your assistant's configuration file.
GitHub MCP Server on GitHub
- Stars
- 34,366on Nov 10, 2026
- Downloads
- 1,514,17266,979 this week
- Growth
- +137stars in 6 days
Stars over time
Facts
- Runs on
- Windows, macOS, Linux
- You need to install
- Nothing extra
- To build from source
- Docker, Go
- Disk space
- 9 MB–27 MB · from the installer
- Accounts or keys
- Your GitHub account, either by browser login (OAuth) or a Personal Access Token
- Cost
- Free · The program is free (MIT) and GitHub's hosted version costs nothing extra. Some of the optional tool groups reach paid GitHub features such as Copilot and the advanced security products.
- License
- MIT (Permissive)
- Last change on GitHub
- Nov 9, 2026
GitHub MCP Server risk signals (5)
- Attentionai can change things in your account
README: "Create, update, and manage issues and pull requests"; the tool list includes repository deletion ("Repository deletion still requires its existing confirmation flow") and a `governance` toolset for "managing rulesets and custom properties at the repository, organization, and enterprise levels". The README itself says to "enable the permissions that you feel comfortable granting your AI tools", and there is a `--read-only` flag for people who want no writes.
AI review
- Attentionassistant reads text written by strangers
The toolsets include `issues`, `discussions`, `notifications` and `pull_requests`, and the README suggests letting "AI help triage bugs". Issue and comment text comes from anyone on the internet, and the same assistant holds the keys to your account, so instructions hidden in that text are a real concern.
AI review
- Attentionsome apps make you paste the token in a file
README, "Handling PATs Securely": "Environment variable support varies by host app and IDE. Some applications (like Windsurf) require hardcoded tokens in config files." The advice given is to run "chmod 600 ~/.your-app/config.json", so the token sits in a plain file on your disk.
AI review
- Infologin credentials built into the official image
Dockerfile comment: "OAuth credentials are injected via build secrets so they are not baked into image history; the values are public in practice but kept out of layers", and the README says "on github.com the official image already includes the app credentials". So the convenient login works only with GitHub's own builds, and a copy you build yourself needs your own OAuth app.
AI review
- Infoexperimental tools behind a flag
README Insiders Mode: "--insiders" or "GITHUB_INSIDERS=true", and for the hosted version the url "https://api.githubcopilot.com/mcp/insiders", described as "early access to new features and experimental tools".
AI review
Evidence is shown in its original language.
What Radar could not check about GitHub MCP Server
- What the hosted version at api.githubcopilot.com keeps or logs, which is not described in this repository.
- Whether the published binaries and the ghcr.io image are built from exactly this code.
- How much of your repository content an assistant pulls in during one conversation.
- Whether a limited token really blocks every write tool, which can only be seen in use.
- Which tools work against GitHub Enterprise Server, where the hosted version is not available.