1utkarsh1/mcp-stdio-guard

mcp-stdio-guard is a command line tool for developers building MCP servers, which are backends that let AI assistants like Claude use external tools. Many MCP servers talk to the AI client over stdio, meaning they exchange messages through standard input and output instead of a network connection. That channel is fragile: a single stray debug message, an accidental print statement, or a startup banner printed by a dependency can corrupt the message stream and make the client fail in confusing, hard to trace ways. This tool catches those problems before a real client does. It starts your server as a subprocess, performs the actual MCP handshake that a client would perform, and can optionally send a real follow up request such as listing available tools. While doing this it checks that every message on stdout is a properly formed message and nothing else has leaked into that channel. It can also scan your source code directly for risky calls like console.log or Python's print that are likely to pollute stdout, and can run the same check multiple times in a row to catch problems that only show up on a second, warm startup. Results are reported as clear pass or fail checks covering the handshake, stdout cleanliness, message formatting, your server's response to the test request, and process behavior such as crashes or timeouts. There is also a machine readable JSON output mode intended for continuous integration pipelines or for tool registries that want to automatically verify a server before listing it, complete with structured issue codes and severity levels. The project is written in JavaScript for Node.js version 18 or later, has no runtime dependencies of its own, and is tested against real servers built with the official MCP SDK. It installs and runs directly with npx, or can be cloned and run from source. It is released under the MIT license.