MCP server for bug fixing

MCP (Model Context Protocol) lets your coding agent call external tools. A bug-fixing MCP server gives the agent eyes on production: instead of you copying error text into chat, the agent queries live bug reports, pulls the diagnosis and code context for each one, and dispatches fixes — all from inside Cursor, Claude Code, or Codex.

Mushi Mushi ships one. No second LLM key required — your editor’s model does the reasoning; Mushi supplies the tools and the data.

What the agent can do with it

List and search reports — “what’s broken in prod?” returns the open queue with severity and plain-English causes.
Get fix context — the files, the diagnosis, similar past bugs, and the blast radius of a change.
Dispatch and submit fixes — hand a fix back to Mushi, which can open a draft PR once GitHub is connected.
Query lessons — past fixes become rules the agent checks before repeating a mistake.

Setup

One command registers the server in your editor’s MCP config:


npx mushi-mushi setup --ide cursor   # or --ide claude

Then ask your agent something like “check Mushi for open bugs and fix the worst one.” The full tool list and configuration options live in the MCP server quickstart.

Hosted or self-hosted

Point the server at Mushi Cloud (50 diagnoses a month free) or at your own self-hosted stack — the MCP surface is identical. Reports come from the SDKs your users already trigger; install those with:


npx mushi-mushi

Next: MCP server quickstart · Connect your editor · Debug Claude Code apps