Mcpsnoop MCP Server
Monitor live JSON-RPC traffic between AI clients and MCP servers in the terminal.
Data last scanned today · Reviewed today
Overview
Mcpsnoop acts as a transparent proxy for MCP communication, sitting directly in the data path between your AI client and server. Unlike testing tools that act as their own clients, this allows you to observe the exact messages and tool calls being exchanged in real-time, aiding in debugging silent failures or incorrect arguments. The tool is written in Go and supports terminal-based inspection, as well as a GitHub Action for CI environments. You can wrap existing server commands or use built-in CLI commands to automatically update Claude Desktop configuration files with the necessary proxy settings.
Our verdict
With a security score of 45/100, this tool carries a high risk level for production environments. This score reflects the fact that authentication methods and read-only mode support have not yet been editorially reviewed. It is suitable for local debugging and development use by those who understand the implications of proxying potentially sensitive MCP traffic, but it should be avoided in sensitive or untrusted production workflows until its security model is audited.
- Zero direct dependencies reduce the supply-chain attack surface
- Provides a live view of JSON-RPC frames for debugging
- Nineteen contributors demonstrate a healthy community effort
- Authentication mechanisms for the proxy remain unreviewed
- Read-only capabilities for the proxy lack formal verification
- Maintained by community contributors rather than the official vendor
Compatibility
| Client | Local | Docker | Remote | Read-only |
|---|---|---|---|---|
| Claude Desktop | ||||
| Cursor | ||||
| VS Code | ||||
| Windsurf | ||||
| ChatGPT |
Frequently asked questions
›How do I integrate mcpsnoop with my existing MCP server configuration?
You can wrap your server command by adding mcpsnoop as a prefix in your MCP config under the command field. Everything after the -- flag should be the original command used to launch your server.
›Does the Claude Desktop helper tool modify my configuration file safely?
Yes, the wrap command creates a backup of your configuration file at claude_desktop_config.json.mcpsnoop.bak before modifying it. You must restart Claude Desktop after running wrap or unwrap for changes to take effect.
›Can mcpsnoop be used for streaming HTTP-based MCP servers?
Yes, you can run mcpsnoop as a reverse proxy using the http command by specifying your target URL and a listen address. This allows you to monitor HTTP status codes and headers in addition to JSON-RPC messages.
›Are Prometheus metrics enabled by default?
No, metrics are disabled by default. You must provide the --metrics-listen flag to start a headless hub that exposes Prometheus metrics at the specified address.
›Is the Prometheus metrics endpoint authenticated?
The metrics endpoint has no authentication. It is recommended to bind the listener to 127.0.0.1, as binding to a network interface will publish tool names to anyone who can reach the address.
›What permissions are required for the mcpsnoop GitHub Action?
The GitHub Action requires security-events: write and contents: read permissions to report findings as code scanning alerts.
Changelog
Badge
Maintain this server? Add the live badge to your README.