About MCProbe

MCProbe was built by Ali Tiknazoglu and was named one of the winners of the CySic AI Hackathon, CyOps Arena.

The Model Context Protocol is how AI agents reach external tools, and new MCP servers are appearing every week. But most ship with tool schemas an agent can call — not ones it can call correctly. Descriptions are missing, parameters are untyped, and, worst of all, some tools silently accept malformed input and hand back plausible-looking garbage instead of an error.

Static linters can flag the obvious smells, but they can't tell you whether a server actually behaves. MCProbe's idea is to go further: it connects to a server, lints its schemas, and then behaviorally tests the tools — calling each one with missing, wrong-typed and out-of-bounds inputs to catch the silent failures static checks can never see. The result is a single 0–100 conformance score, an A–F grade and a prioritized list of exactly what to fix.

The audit engine is open source and also runs as an MCP server and a command-line tool. This site is the hosted version: paste a URL to audit any remote MCP server in your browser or audit a local (stdio) server from the CLI with your API key — the report lands in your account either way. Pro adds 30 audits a day, the full report, local-server audits and the public gallery.