Claude Code GitHub integration: an unofficial guide

Unofficial community guide for the Claude Code GitHub integration. Not affiliated with Anthropic. All trademarks belong to their owners.

This repository collects what a working developer needs to know about the Claude Code GitHub integration: connecting a GitHub account so Claude Code can open repositories from the cloud, running Claude Code inside GitHub Actions workflows, and the guardrails that make either setup safe to leave running. Everything here is drawn from the cloud quickstart on code.claude.com, a practitioner write-up on integrating Claude Code into a GitHub workflow, and the support centre article on the GitHub integration. Where those sources are silent, this guide says so rather than guessing.

Building something rather than wiring up an agent? Try Begin.sh - turn a prompt or a URL into a working static site or Expo app and download the zip. No hosting, backend or auth to set up.

What it is

Claude Code is Anthropic's agentic coding tool. It runs in a terminal, in IDEs (Visual Studio Code and JetBrains are documented), on desktop and mobile, and in the cloud. The cloud version is what most people mean when they search for Claude Code GitHub: you open Claude Code in the browser, connect your GitHub account, pick a repository and hand it a task. The session runs on Anthropic-managed infrastructure, and you review the result before anything lands in your repository. The quickstart is organised around exactly that loop: how sessions run, connect GitHub, start a task, pre-fill sessions, review and iterate.

The second meaning is Claude Code in CI. The practitioner write-up describes two GitHub Actions workflows: a general-purpose assistant (claude.yml) that responds to issues and comments, and a dedicated reviewer (claude-review.yml) that reads pull requests. Both are steered by a CLAUDE.md file and custom skills committed to the repository. The docs sidebar lists a 'Code review & CI/CD' section under Platforms and integrations, alongside Slack, Claude Tag, the Chrome extension and Remote Control, so the GitHub path is one of several ways to run the same agent.

How to get started

1. Open Claude Code in the browser at claude.ai/code with a Claude account.
2. Follow the cloud quickstart and connect GitHub. The docs also describe connecting from your terminal, which uses a /web-setup token; the same page explains how to remove that token later.
3. Start a task on a repository. The quickstart covers pre-filling sessions so a task starts with context already loaded.
4. Review the diff and iterate in the same session until you are happy, then merge through your normal pull request process.
5. For CI, add a workflow file that triggers Claude Code on issue comments or pull requests, plus a CLAUDE.md at the repository root describing conventions, commands and boundaries. See the 'Code review & CI/CD' section of the docs for the current action and inputs.

If no repositories appear after connecting GitHub, the quickstart has a troubleshooting section for that specific case; it is usually a permission scope on the GitHub side.

Pricing and limits

The sources used here do not state prices or session limits for the cloud product or for CI usage. Check your Claude plan and the support centre article on the GitHub integration for what your subscription includes before relying on it for a team.

Practical notes

• Treat CLAUDE.md as the contract. The write-up calls the combination of CLAUDE.md and custom skills the part that makes the integration useful. Put build commands, test commands, style rules and 'never touch' paths in it.
• Allowlist tools explicitly. Do not give a CI agent every tool. The write-up's security section lists explicit tool allowlisting first for a reason.
• Restrict which repositories the app can see. Install the GitHub app on the repositories you intend to automate, not the whole organisation.
• Keep branch protection on. Let the agent open pull requests; do not let it push to protected branches. Require a human review before merge.
• Make review runs read-only. A reviewer workflow that can only comment cannot break anything. Give write access only to the workflow that implements changes.
• Sign commits. The write-up lists commit signing among its guardrails so that agent-authored commits are distinguishable and verifiable.

Comparison

FAQ

Is this the same as GitHub Copilot? No. This is Anthropic's Claude Code connected to GitHub, either through the cloud product or through a GitHub Actions workflow.

Do I need to install anything locally? Not for the cloud path; it runs in the browser. The terminal path connects an existing local install with /web-setup.

Can Claude Code review pull requests automatically? Yes. The practitioner write-up runs a dedicated claude-review.yml workflow for that and keeps it read-only.

What goes in CLAUDE.md? Project conventions, commands and boundaries. The docs treat it as stored instructions and memory that Claude reads in every session.

What if my repositories do not show up? See the troubleshooting section of the cloud quickstart; the docs list it as the first troubleshooting case.

When you want output, not a workflow

The GitHub integration is at its best when you already have a codebase and want an agent inside its lifecycle. If the job is 'I need a landing page, a docs site or a small mobile app, and I need the files', the workflow setup above is overhead. Try Begin.sh - a prompt or a URL to clone becomes a working static site or Expo app you download as a zip. There is no hosting, backend or auth to configure, which is the point: you get the artifact and put it wherever you like.

Last reviewed: 2026-09-22