There are already good PR review tools — CodeRabbit, Copilot Code Review. So why did we build claude-pr-reviewer?
The Problem with Existing Tools
CodeRabbit is polished and works well. Copilot is... also fine. But both have friction:
You don't choose the model. CodeRabbit runs their own AI. Copilot runs Claude 3.5 Sonnet. You're locked in. If you want cheaper reviews (Haiku at $0.001/PR) or better ones (Opus at $0.01/PR), you can't.
They don't post as GitHub reviews. They comment on the PR, sure. But not as a formal review that shows up in the review queue. No "REQUEST CHANGES" verdict. No structured pass/fail that blocks merges.
The API exists but is hard to use. GitHub's PR Reviews API lets you post inline comments on specific lines with proper formatting. Most tools don't actually use it. We do.
You're paying them rent. CodeRabbit charges per org or per seat. Fine if you have budget. But if you just want cheap, fast code review on your side projects? $19-50/month is friction.
What claude-pr-reviewer Actually Does
- Posts as a GitHub Review. Not a comment. A real review with "REQUEST CHANGES" or "COMMENT" verdict. Blocks merges if configured. Shows in the review queue.
- Inline comments on problem lines. Click the line number, see exactly what Claude flagged and why.
- You pick the model. Haiku ($0.001/PR), Sonnet ($0.003/PR), Opus ($0.01/PR). Fast and cheap by default. Upgrade for precision.
- Actually catches stuff. We tested it against CodeRabbit reviews side-by-side. Same issues found, but we found a few more because we're more thorough on code path analysis.
- GitHub Action (zero setup). Add 4 lines to your workflow file. Every PR gets reviewed automatically.
-
Or use the CLI.
python pr_reviewer.py <url>anywhere, anytime. - Configurable strictness. Lenient (security + bugs only), Balanced (standard), Strict (includes perf, tech debt, missing tests).
The Math
A typical code review on CodeRabbit or Copilot costs you $19-50/month minimum, or $0 if you're already paying for the enterprise subscription.
With claude-pr-reviewer running as a GitHub Action on Sonnet:
- 10 PRs/day = $0.03/day = ~$1/month
- 50 PRs/day = $0.15/day = ~$5/month
- 100 PRs/day = $0.30/day = ~$9/month
Even if you're on a free tier, that's cheaper than CodeRabbit's entry price. And you're not paying for seats, orgs, or minimum monthly fees.
For teams: you own the GitHub Action. No vendor lock-in. No surprise price increases. You can switch models mid-stream if Anthropic releases something better.
What It Doesn't Do
- Real-time feedback. CodeRabbit can comment as you draft. We post reviews after the PR is created.
- Conversation. This is one-way feedback, not chat. Good enough for most teams.
- Integration with other tools. We post to GitHub. That's it. You want Slack notifications? Wire it yourself; GitHub Actions has that built-in.
- Fancy dashboard. CodeRabbit has metrics, trends, team comparisons. We have... a comment on your PR. Use GitHub's built-in insights if you care about metrics.
Why It Exists
I built this because every few months I'd see a request in dev.to comments or GitHub issues: "Why doesn't CodeRabbit let me use a cheaper model?" "Why can't I use my own API key?" and "Is there a self-hosted option?"
The answer to all three was no. So I spent a few hours building something that said yes.
It's also a case study in "when you don't need to build something, build it right anyway." CodeRabbit does 80% of what anyone needs. We do that 80% as a GitHub Action you can modify, with model choice you control, for pennies instead of dollars.
Getting Started
- uses: indoor47/claude-pr-reviewer@v1
with:
anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
Top comments (0)