This is a quality verification layer that tests MCP servers and AI agents before you trust them with real work. It exposes tools like check_quality() and verify_attestation() through the MCP protocol, runs adversarial probes for prompt injection and PII leakage, and scores targets across six dimensions weighted by accuracy, safety, and reliability. Behind the scenes it uses a battle arena with Bayesian ELO ratings and adaptive IRT testing to calibrate questions, then issues W3C Verifiable Credentials as cryptographic proof. Reach for this when you need pre-deployment competency gates rather than post-hoc reputation scores, especially if you're building agent marketplaces or payment-gated services that need quality attestations before transactions.
claude mcp add --transport stdio io.github.vitaliiserbynassisterr-agenttrust uvx agenttrust