Summary

It tells you how the scores are made, which is rarer than the scores.

The ranking tool states its own method up front: liveness derived from HEAD probes, three samples from a single host, not a rigorous benchmark. Read it as a signal that a provider is up and answering, not as a quality measure. The receipt tools are the sturdier half — offline verification against a receipt envelope you already hold does not depend on trusting the service at all.

What it is

A read-only client for TrustBench, an audit layer for x402 agent payments. It ranks providers by capability and verifies the Ed25519-signed receipts that record what was paid, to whom, and for what.

What you get

A scored provider list for a capability — search, inference, data, media or infra — with latency and success-rate telemetry — get_rankings
Any routing receipt fetched by its id, as an immutable signed record of a routing or payment event including its on-chain settlement reference — get_receipt
Signature verification in two modes: by id, where the server fetches the receipt and re-runs the check, or offline against a receipt JSON envelope you already hold — verify_receipt

Requirements

No API key and no account — npx -y @trustbench/mcp and restart the host. TRUSTBENCH_BASE_URL overrides the API base for local development. All three tools read; nothing here spends or moves anything.

Setup effort

One command — npx -y @trustbench/mcp