Making PR comments useful again

I build software, I ship a lot, and I like tools that stay out of the way. When I first tried CodeRabbit I had that “oh wow” moment — then two days later I had 20+ comments across a PR and the vibe switched from “helpful” to “now I’m triaging an inbox.” Some comments were great, some were meh, some were just confidently wrong. Either way, I still had to read everything, validate it, fix stuff, write a reply, resolve the thread, and keep going. Rinse, repeat.

So I did what we all do: scratched the itch. I built a tiny MCP server so I could tell my AI assistant (Cursor, Claude Code, etc.):

It fetches unresolved comments (CodeRabbit or human), groups them by severity, explains what each one means in context, tells me if it’s legit, proposes a diff, and — if I approve — applies the change locally, replies on GitHub, marks the thread resolved, and moves on to the next one. Basically: sanity for the “read → validate → fix → reply/resolve” loop.

Here’s a quick feel for the workflow:

The first version was intentionally boring: pull comments, validate with project context, propose a fix, close the loop. Once that felt solid, I started sanding down real‑world edges. You know those lingering PRs with 27 unresolved comments across 6 files? I added smarter helpers: listMyActivePrsWithUnresolvedComments so I can see “what still needs love” at a glance, and bulkResolveOutdatedComments to sweep the zombie threads that refactors made irrelevant. Less tab‑hopping, more progress.

Day to day, the point is flow: it’s fast, it stays out of your way, and your code stays on your machine. Setup takes minutes, and once it’s wired up you can just point it at a PR and start closing loops.

And because nobody ships alone: you can invite teammates and share the same resolve‑and‑reply workflow, so it’s clear what got handled without hunting through chat logs. Per‑repository token isolation is on my roadmap.

This is still early, and I’m shipping almost daily. If you’ve got ideas or sharp edges you want gone, tell me — I actually build from feedback. Easiest way is the contact page.

If you want to try it (free plan, no card), setup takes about five minutes:

1. Sign up and generate an API key.
2. Create a GitHub Personal Access Token (repo scope), then add your repository and token in Settings.
3. Add the MCP server to your assistant. See the quick start docs.
4. Paste the test command above into a new chat.