Thousands of MCP servers are shipping, and nothing tells you if one is safe or does what it claims — tool descriptions can hide prompt-injection, destructive tools ship with no warning, and nothing is provable later. Shield reads any server's tools, grades it for safety and transparency, and seals the exact result into Bitcoin. Free. The AI labs build the protocol; someone independent has to referee the trust. That's us.
Shield checks the signals a reviewer checks: tool annotations (title, read-only / destructive), prompt-injection patterns in tool text, destructive tools shipped without a warning, and description quality. A PASS is a transparency and hygiene signal, not a guarantee. Every result is sealed and public. Browse the Verified-MCP registry →