How to debug an app Cursor wrote for you

You shipped an app with Cursor. A user says “the save button doesn’t work.” Now what? You didn’t hand-write the code, so the usual debugging loop — reproduce, read the stack, remember why that module exists — starts from zero every time.

Mushi Mushi closes that gap. Users report bugs from inside your app (a widget click or a shake on mobile), and Mushi turns each report into a plain-English read on what broke and why, with a fix prompt scoped to the right files. Your Cursor agent picks it up over MCP and applies the fix without you leaving the editor.

The loop, end to end

1. Install the SDK — npx mushi-mushi detects your framework and wires everything up.
2. Connect Cursor — npx mushi-mushi setup --ide cursor registers the Mushi MCP server, so Cursor can list open reports, pull fix context, and dispatch fixes.
3. A user hits a bug — the report lands with a screenshot, the user’s own words, and Mushi’s diagnosis of the cause.
4. Ask Cursor “what’s broken in prod?” — it reads the queue, proposes the fix, and you merge. Repeat reports about the same bug collapse into one row, and each fix becomes a lesson Cursor sees on future PRs (.mushi/lessons.json).

Why not just paste the bug into chat?

You can — but a user’s “it’s broken” message has no repro, no console output, and no file pointers. Mushi’s report carries the screenshot, session context, and a diagnosis grounded in your actual repo, so the agent starts from the answer instead of the question.

Set it up

npx mushi-mushi
npx mushi-mushi setup --ide cursor

Free tier: 50 diagnoses a month, no card. Open source, self-hostable.

Next: Cursor integration guide · MCP server quickstart · Incident loop walkthrough