Code Review
Find bugs at the pace of agentic code
Claude automatically reviews PRs, verifies findings, and posts comments ranked by severity, so you can catch bugs before they reach production.
Available on Claude Team and Enterprise plans with repos on GitHub. Compare plans.
Coding agents help teams write more code than reviewers have time to read closely. When pull requests pile up, a quick read can miss race conditions, security flaws, and other issues. Code Review reads each change closely first to help catch these before you have merged.
Find bugs beyond the diff
Code Review reads each change and the code around it to help catch bugs a pull request exposes but didn’t cause.
Verify findings before they post
Bugs Claude finds are checked against your code before posting, so reviewers get findings worth their time.
Rank bugs above nits
Findings are ranked by severity and tagged Important, Nit, or Pre-existing, so serious ones are shown and can be actioned first.
Post findings as inline comments
Findings are shared as inline comments with a summary. With reviews on every push, comments resolve once you fix them.
Focus on what breaks production
By default, reviews look for bugs that would break production. Your team can change what gets flagged in REVIEW.md.
Get findings while the change is fresh
Most reviews finish in 6 to 10 minutes, so findings are waiting by the time a reviewer opens the pull request.
Run a review from the terminal
If you’re not an admin or not on Team or Enterprise yet, you can still run a review yourself. Ultrareview is a deeper review you start from Claude Code with the command /code-review ultra, on a branch or a github.com pull request. Reviewer agents look for bugs in a cloud sandbox while you keep coding, and findings are verified before you see them. Ultrareview is in research preview, and paid runs need usage credits turned on.
Your team sets the rules, and merging stays with you.
Pick where reviews run
Run a review when a pull request opens, on every push, or only when someone asks. New repos start on every push.
Choose repos and a review level
Pick a repo you want to review. Each starts on Standard, or set your own level: Quick for routine code, Deep for what’s critical.
We built Code Review to keep up with the code Claude writes. After months of daily use, it’s one part of how we handle more volume with more coverage and better quality. Our advice is to let it take a first pass on each change, so your reviewers can focus on the riskiest code. Here are the numbers from our own repos in 2026.
30,000+ changes merged a day
As of September, with Anthropic developers writing more of their code with Claude. Up from about 2,500 in May.
54% get substantive comments
Up from 16% of our pull requests before Code Review. That’s more than three times the share, as of our research preview.
40% drop in Claude Code revert rate
From June to September, merged pull requests more than doubled while the share we had to revert fell.
Code Review is an AI code review tool that reads each GitHub pull request and posts the bugs it finds as comments for reviewers. It allows Claude, the AI technology behind Code Review, to take a first pass at code reviews. Code Review uses AI and may miss bugs, security vulnerabilities, or other issues.
Code Review works with repos on GitHub, including GitHub Enterprise Server and GitHub Enterprise Cloud with data residency. It doesn’t support GitLab, per-user access controls, or self-hosted environments. It isn’t available to organizations with zero data retention, the HIPAA configuration, or customer-managed encryption keys.
No. Code Review runs on your GitHub pull requests and doesn’t require Claude Code.
Yes. Run /code-review in a Claude Code session on any plan to review your own branch. Reviews on every pull request need a Team or Enterprise plan.
Yes, with a Claude Team or Enterprise organization. Reviews run on Anthropic infrastructure and are billed by Anthropic.
Inside a review, access to your repo is read-only and nothing is pushed. The Claude GitHub App you install holds broader permissions that other Claude features share. A review reads the diff and the surrounding code without building or running it, then posts comments and a check run on the pull request. When a later push fixes a finding, the review resolves its own comment.
Setup has two steps, one in Claude and one in GitHub. In Claude, an organization owner should turn on Code Review in admin settings. In GitHub, an organization owner should install the Claude GitHub app. Next, choose which repos Code Review can review. If you don’t have owner access in either, send your contact the setup guide.
No. Code Review picks the models. To make reviews more or less thorough, change the review level.
Code Review is billed on token usage through usage credits, separately from your plan’s included usage. Admins can set a monthly spend limit and see the average cost per review for each repo.
Try it on one repo, then add more.
Ask for a review in a comment
Anyone with write access can comment
@claude reviewon a pull request to start a review, drafts and forks included.