Beta. The policy is v0.2 and the checks are still being tuned. Treat a grade as a reason to evaluate further.
Should you connect
that MCP server?
Paste a server. Get a grade against a published security policy - provenance, capabilities, transport, known vulnerabilities, and tool-poisoning patterns. Every check is open source.
try
Browse the directory – pre-scored grades for every MCP server
| Check | Policy | Method |
|---|---|---|
| Provenance | §1 | Official registry listing, repo health, package ↔ repo consistency, provenance attestation |
| Capabilities | §2 | Tool surface via tools/list or static source scan - never executed. Shell, filesystem, egress, and credential access flagged |
| Transport & auth | §3 | HTTPS, auth required, OAuth 2.1 resource metadata (June 2025 spec) |
| Vulnerabilities | §4 | OSV.dev advisory lookup |
| Instruction integrity | §5 | Hidden instructions, zero-width characters, cross-tool shadowing, and authority over-reach across tool descriptions, the server's initialize instructions, prompts, and resources; schema-change (rug-pull) detection across scans |