Connect GitHub to Claude Code

Let Claude open issues, comment on pull requests, and push fixes in GitHub. From the terminal, inside the sessions where you already work.

Build my connector Talk to us

How do you connect GitHub to Claude Code?

Claude Code reaches outside tools through MCP servers. To connect GitHub, you add a GitHub MCP server: a small hosted service that exposes GitHub's data and actions over the Model Context Protocol. With a custom connector we build for you, that is one command in the terminal, then an OAuth sign-in with your own GitHub account. No API keys pasted into config files.

claude mcp add --transport http github https://your-connector-url/mcp

Then run /mcp inside a Claude Code session to sign in. Add it with --scope project and the connector is checked into the repo for your whole team.

What can Claude Code do in GitHub once connected?

Full read-write GitHub access lets Claude open an issue from a bug report, comment on a pull request, push a fix to a feature branch, label and assign work, and request review once you approve. The triage and follow-up that used to mean tab-switching happen where the code already lives.

Open an issue from a bug report, with labels and an assignee
Comment on a pull request or leave a review summary
Push a change to a feature branch and open a draft pull request
Apply labels, assign reviewers, and move issues across a project board
Close or reopen issues as you triage a backlog
Request review on a pull request once you approve

One connector, every Claude surface

The connector we build is a remote MCP server, which is the same standard claude.ai and Claude Desktop use. Add it once and the same GitHub access works in Claude Code for you, in claude.ai for the rest of the team, and in Claude Desktop. One build, one set of permissions, one monthly plan covering all of them.

Build it yourself, or have it built?

If you live in the terminal, you can absolutely write your own MCP server against the GitHub API. The protocol is open. What you take on is everything around the happy path: OAuth token refresh, hosting, monitoring, keeping up when GitHub changes its API and when the MCP spec itself moves, and scoping write actions so an agent can never do more than you intended. We build GitHub connectors with separated read and write tools and approval gates on destructive actions, then host and maintain them so they keep working while you work.

Frequently asked questions

Does the same GitHub connector work in claude.ai and Claude Code?

Yes. It is one remote MCP server. Claude Code adds it with a single "claude mcp add" command; claude.ai and Claude Desktop add it as a custom connector in Settings. Each person signs in with their own GitHub account, so access always matches what that person can already see and do.

Do I need to host or maintain anything myself?

No. The connector runs as a hosted service we operate: we handle uptime, token health, API changes on GitHub's side, and MCP spec migrations under the monthly plan. Your side is one command in the terminal and an OAuth sign-in.

Is it safe to give a coding agent write access to GitHub?

Write access is scoped, not blanket. Read tools and write tools are separated, destructive actions like deleting or sending are gated behind your explicit approval, and the connector can never exceed the permissions of the GitHub account it signs in with. Revoking access is one click in GitHub.

More on the GitHub connector

GitHub connector for claude.ai GitLab + Claude Code PagerDuty + Claude Code

Tell us what you run. We'll build the connector.

Pick your tool, choose what Claude should do, and get an instant scope and price. No call required.

Build my connector Talk to us