What does connecting a coding agent do?
Your coding agent reads Rapid Review’s review of its pull request, fixes what the review found, pushes, and waits for the next review, until the pull request is ready. You don’t pass comments or prompts between them. At the end, your agent tells you one line:- Ready to merge: nothing is open. The line says what was fixed, and names any finding your agent declined as not a defect, so you can look at those.
- Stuck: findings are still open after 3 reviews, the account is out of credits, or a review didn’t run. Tell your agent to go on, or look at the pull request.
How do I connect my agent?
Add Rapid Review’s server, https://app.rapidreview.dev/mcp{/* /fact */}, to your editor. The first time your agent uses it, it opens the Rapid Review app, where you sign in with GitHub and allow the agent. Connected agents in the app shows the same setup lines. Claude Code:~/.cursor/mcp.json, or .cursor/mcp.json in a project:
How do I start it?
After your agent pushes to a pull request, ask it to get the pull request ready with Rapid Review. In Claude Code and Cursor you can also use the commandfinish_review, with the pull request, or with none for the branch you’re on. In Claude Code it’s /mcp__rapid-review__finish_review.
Rapid Review’s server also tells agents that read its instructions, such as Claude Code and Codex, to start after each push.
What can my agent do as me?
On the consent page, you allow an agent to:- read Rapid Review’s reviews of pull requests in the repositories you can see in the app;
- reply in a finding’s thread as you, to decline it as not a defect. Each reply ends with “Sent by a coding agent through Rapid Review’s review loop.”;
- ask for a review of a pull request, at most 5 times a day on each. Each review uses credits, as any review does.