# kensaurus/mushi-mushi [Health: Active]

**Category:** 🗄️ Databases  
**Repository:** https://github.com/kensaurus/mushi-mushi  
**GitHub Stars:** 3  
**npm Downloads (last month):** 1329  
**Views:** 2  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/kensaurus-mushi-mushi

## Description
Bug-report triage and fix-dispatch MCP for vibe coders. Ask your AI coding agent for a plain-English diagnosis of any user-reported bug — with evidence, blast radius, and a paste-ready fix prompt — then dispatch an automated GitHub fix PR without a second LLM key. 80+ tools, Supabase-backed, MIT SDK. npx @mushi-mushi/mcp

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "mushi-mushi": {
    "command": "npx",
    "args": ["-y","mushi-mushi"]
  }
}
```

## Documentation

## What kensaurus/mushi-mushi does

kensaurus/mushi-mushi exposes Mushi Mushi's bug-triage and fix-dispatch workflow through MCP. It is designed for applications that collect reports from users and need an agent to turn those reports into actionable engineering work.

Reports can include a screenshot, the affected route, the user's note, and recent console and network activity. Mushi Mushi groups duplicate reports, produces a plain-English diagnosis, and supplies evidence, an assessment of impact, and a prompt that a coding agent can use to make the change. The queue can also combine reports from application users with Sentry errors, store reviews, and Slack submissions.

The server is aimed at workflows using Cursor or Claude Code, including projects built with AI-assisted development tools such as Lovable and Bolt. It is useful when a bug does not produce a conventional exception, such as an obscured button, an action that appears unresponsive, or a slow dashboard.

## How it works

A Mushi widget captures a report from the application. The system records the available client context, consolidates duplicate reports, and generates a diagnosis. An editor or agent reads that diagnosis through MCP and can use it to implement a fix.

Fixes can be dispatched as draft GitHub pull requests for human review. The hosted fixer does not require a separate key from the user, while Cursor agents use the user's Cursor credentials. Fixes can also run on the user's computer with the Mushi Mushi worker. When configured with an Anthropic key, a second AI reviews an agent pull request and recommends merging, making changes first, or holding it. A held pull request is not merged unless someone overrides the decision in the console and supplies a reason.

## Setup and configuration

The command-line setup starts with:

```bash
npx mushi-mushi
```

The wizard detects the framework, installs the relevant SDK, writes environment values, and sends a test report. After that, connect the editor from the Mushi connection page or run `npx mushi-mushi setup --ide cursor` or `--ide claude`. The project also provides an MCP package for editor integration and a CLI for setup, reports, fixes, and health checks.

Application SDK configuration uses a project ID and API key. Vite projects use `VITE_MUSHI_PROJECT_ID` and `VITE_MUSHI_API_KEY`; Next.js projects use the corresponding `NEXT_PUBLIC_MUSHI_*` variables. Framework packages are available for React, Vue, Svelte, Angular, React Native, Capacitor, and framework-independent web applications. Native iOS, Android, and Flutter SDKs are preview packages.

A hosted free plan supports one app and 100 diagnoses per month. The backend can also be self-hosted; its implementation uses Supabase edge functions and Postgres.

## Tools and capabilities

- Retrieve and inspect Mushi bug reports through MCP.
- Read plain-English diagnoses with evidence and likely root causes.
- Review a report's blast radius and related duplicate reports.
- Produce paste-ready prompts for coding agents.
- Dispatch automated GitHub fixes as draft pull requests.
- Connect reports and fixes to an editor-based development workflow.
- Run setup, reporting, fix, and health checks through the CLI.

## Limitations and notes

The MCP workflow depends on application reports being captured through a Mushi SDK or another supported intake path. The widget and SDK require project credentials. Hosted fixing avoids a user-supplied LLM key, but Cursor agents require the user's Cursor key; optional second-AI pull-request review requires the user's own Anthropic key. Draft pull requests remain subject to human review, and the server does not automatically merge a pull request that its review marks for holding.

_Full upstream README: https://allmcps.com/mcp/kensaurus-mushi-mushi/readme_