Introducing MergeMitra
Most AI PR tools either flood the diff with noise or drift into writing the code for you.
MergeMitra is built for a different job: catch the issues that matter early, package them cleanly, and make follow-up reviews smarter.
It catches the stuff teams actually miss
MergeMitra is strongest when a pull request looks fine at first glance but still hides real problems.
Edge cases. Broken assumptions. Weak validation. Missing tests. Accessibility misses. Design choices that will make the next change harder.
We built it on GPT-5.2 because review is a long-running task. The model has to stay consistent across larger diffs, grouped findings, and follow-up passes, not just produce a clever one-off answer.
We also use MergeMitra on our own pull requests. That has improved our codebases in practical ways: tighter tests, clearer PR context, better follow-through on fixes, and fewer issues slipping through just because the first review looked done.
It also taught us a product lesson we care about a lot: a review agent should not behave like a coding agent.
One comment to fix, one comment to review
MergeMitra keeps issue-finding and reviewer guidance separate on purpose.
The main PR Review comment is where actionable issues live. Same root issues are grouped across files, locations are attached, and smaller nitpicks are collapsed so the important problems stay visible.
That makes the output easy for humans to scan and easy for another LLM or coding agent to use when it is time to fix things.
Then there are arch notes.
Arch notes are a separate comment for the human reviewer. They carry the PR Overview, identify the kind of change being made, and call out the Focus Areas for Architect Review.
They are not just a second copy of the issue list. They help a reviewer decide where to spend time.
Minimal inline suggestions, by design
Inline suggestions are intentionally optional and minimal.
You can turn them off entirely, allow all of them, or keep only the top N. That is deliberate. MergeMitra is here to point out issues, not quietly take over the PR.
We think that separation matters. A coding agent can help implement fixes. A review agent should stay focused on finding problems, preserving context, and guiding judgment.
Rereviews that remember the last conversation
MergeMitra does not treat every follow-up pass like a brand-new PR.
When you ask for a rereview, it analyzes the changes since the last reviewed commit and keeps previous bot and reviewer conversation in context.
That means follow-up reviews can clearly show what was fixed, what was partially fixed, and what is still unresolved, instead of starting from zero each time.
And if there are no new reviewable changes, MergeMitra says that instead of posting another noisy review.
How to use MergeMitra
Install the GitHub App, connect your repositories, and leave auto review on if you want reviews when a PR is opened or when a draft PR becomes ready for review.
You can also trigger reviews manually:
Ask for an incremental pass after pushing fixes:
See the available commands:
Aliases work too:
You can use MergeMitra without any repo config at all.
If a repository needs its own behavior, you can optionally add a .mergemitra.json file at the repo root:
That lets you bring in your own review guidelines, validate against your PR template, filter noisy files, and control how many inline suggestions you want.
MergeMitra is built to catch real issues early, keep the output readable, and make every rereview more useful than the last.