realwigu/mcp-doctor

**Diagnose, secure, and benchmark your MCP servers.**
Zero-config CLI that auto-discovers MCP server configs across Claude Code, Cursor, VS Code, Windsurf, and Claude Desktop — then tests connections, flags security issues, and benchmarks latency in seconds.
Why?
MCP servers are becoming the backbone of AI-assisted development. But as you add more servers across more tools, things break silently:
- **Servers go down** and you don't notice until a tool call fails mid-conversation
- **Secrets leak** — API keys hardcoded in config files, tokens visible in process args
- **Slow servers** drag down your entire AI workflow without you realizing it
- **Configs drift** between tools — what works in Cursor might be broken in Claude Desktop
mcp-doctor gives you a single command to check everything, across every tool, in seconds.
Quick Start
That's it. No config needed — it finds your servers automatically.
Commands
| Command | Description |
| ------- | ----------- |
| doctor | Run all checks at once (scan + security + bench) |
| scan | Test all MCP server connections |
| security | Audit configs for security issues |
| bench | Benchmark server response times |
| serve | Run as an MCP server (stdio transport) |
All commands support --json for machine-readable output.
doctor — Full checkup (recommended)
Runs scan, security, and bench in one go and prints a summary.
scan — Test all MCP server connections
Discovers configs and verifies each server responds to a JSON-RPC handshake.
security — Audit configs for security issues
Checks for leaked secrets, overly broad permissions, and risky command patterns.
bench — Benchmark server response times
Measures JSON-RPC round-trip latency for every configured server.
MCP Server Mode
mcp-doctor can also run **as an MCP server itself**, exposing scan, security, bench, and doctor as tools your AI assistant can call directly.
When invoked without arguments and stdin is piped, it automatically starts in server mode using stdio transport. You can also explicitly run:
This means your AI assistant can diagnose its own MCP infrastructure on demand.
GitHub Action
Use mcp-doctor in CI to catch broken servers and leaked secrets automatically:
The action outputs JSON via ${{ steps.mcp-doctor.outputs.result }} for downstream processing.
JSON Output
All commands support --json for structured output — useful for CI pipelines, dashboards, or scripting:
Supported Tools
| Tool | Config Auto-Detected |
| --------------- | -------------------- |