Vouch

Gives Claude a trust score API before it sends money to a counterparty. Exposes two MCP tools: vouch_score checks a target URL against threat feeds, transport signals, domain heuristics, and crowdsourced reputation data accumulated in D1, returning an explainable risk level. vouch_report lets you flag or vouch for hosts to feed the reputation layer. The scoring engine uses weighted signals with a safety override so a single hard negative can't be averaged away by softer positives. Runs on Cloudflare Workers and charges per call via x402 microtransactions in USDC on Base mainnet. If you're building an agent that autonomously pays merchants or APIs, this is the scam screen you'd call first.