AgentAvow

Signed, independently-recomputable safety scores for the MCP servers, packages, and tools an agent connects to. Static analysis across 12 categories yields a 0–100 score, per-finding detail, and an Ed25519/JWS attestation anyone can recompute offline against a public JWKS — so a score can't be forged or silently changed. Free, no account. Ships a one-line shields.io README badge, a GitHub Action to gate CI merges on a minimum score, an MCP server (`uvx agentavow-trust`, registry `com.agentavow/agentavow-trust`) so an agent can check "is this safe to install?" before connecting, and watch-alerts when a tool's grade drops or its signed definition changes (rug-pull detection). agentavow.com/check.

Project

Browse this MCP server in the TensorBlock MCP Index.