I am a junior backend developer who wants to start my career now. Here's some info about my project: The MCP Inspector is a manual GUI for poking at a server by hand. mcpcheck is the opposite: a non-interactive audit you can run in CI and get a grade from. And it grades the thing that actually hurts an agent — not spec conformance (the SDKs already handle that), but ergonomics: bloated tool descriptions burn context on every single request 40+ tools wreck tool-selection accuracy a 20k-token tool response can blow up a whole session ogs on stdout silently corrupt the stdio transport mcpcheck connects as a real MCP client (stdio or streamable HTTP), introspects the server, runs static and dynamic checks, and prints a scored report.

mcp-checker

One command, 0–100 score, list of what breaks for the agent.

One command, 0–100 score, list of what breaks for the agent.