Mushi is an open-source bug diagnosis and fix dispatching tool for AI rapid development scenarios. Its core goal is to help developers quickly understand why AI-generated applications fail and to greatly simplify the repair process.
Primary use: When an end user encounters a bug, Mushi can capture the problem context, automatically analyze an easy-to-understand root cause, and provide directly usable fix prompts. It is specially designed to work with AI coding environments such as Cursor, Claude Code, Lovable, or Bolt, reducing the handling time for each bug from hours to minutes.
Main capabilities:
- Capture: When a user clicks a component, it records a screenshot, the current route, user notes, and the latest console and network events.
- Diagnose: Write the root cause in plain English and attach fix prompts. If the same button errors 20 times, it will be merged into one line.
- Fix: The editor can read diagnostic information through MCP, and the agent can automatically open a draft PR for human review.
- Cover non-throwing bugs: For example, a payment button being obscured by the keyboard, a save button not responding to repeated clicks, or a dashboard taking 12 seconds to load.
- Integration capabilities: Supports connecting to Sentry (optional), and can also import Sentry alerts into the same queue via Webhook, with bidirectional synchronization; it also provides adapters for Datadog, Bugsnag, Rollbar, Crashlytics, etc., and plugins can synchronize with Linear, Jira, GitHub Issues, Slack, Discord, Teams, PagerDuty.
How to use:
- The SDK requires two environment variables: VITE_MUSHI_PROJECT_ID / VITE_MUSHI_API_KEY, or NEXT_PUBLIC_MUSHI_* (see examples/sdk.env.example). No Sentry, LLM key, or monitoring stack is required.
- Install Mushi skills in Cursor or Claude Code: run npx skills add kensaurus/mushi-mushi, then run the mushi-setup skill to connect the Mushi SDK and MCP server to the application, and send a test report to verify the connection.
- The skill also provides commands such as /mushi-debug, /mushi-health, /mushi-integration.
- Native SDKs (iOS, Android, Flutter) are in preview and can be installed from packages/ios, packages/android, packages/flutter. A runnable demo is available at examples/react-demo.
- The managed free version includes 50 diagnoses per month with no credit card required; it can also be self-hosted, but self-hosting requires bringing your own LLM key and paying according to the vendor's listed price.
Use cases: Suitable for developers using AI programming tools to rapidly deliver applications, especially teams that want to reduce bug investigation time and turn user feedback into actionable fix steps.
Limitations and notes: The native SDKs are in preview; self-hosting requires bearing LLM costs yourself; the managed service is billed by number of diagnoses and has a spending cap; enterprise-grade integrations (SSO, retention, regional routing, Kubernetes) require referring to the docs/operators/ documentation. Other details are not yet provided.
