x402-safe
v0.1.0
Published
Verify a counterparty before you pay it on x402. 78% of x402 recipients are unverifiable ghosts; this checks AgentRank before your client settles.
Maintainers
Readme
x402-safe
Verify a counterparty before you pay it on x402.
78% of everything paid on x402 is an unverifiable ghost — a wallet exactly one other wallet has
ever paid. Your agent is about to send real USDC to one of them. x402-safe checks
AgentRank at the moment a server answers 402, and blocks the payment if the
recipient is not a real, settlement-backed agent.
It is zero-dependency and client-agnostic: you keep your x402 payment library, x402-safe just gates it.
Install
npm install x402-safeUse
import { safeFetch } from "x402-safe";
// your existing x402 payment executor (x402-fetch, x402-axios, your own)
const pay = (paymentRequirements) => x402Pay(url, paymentRequirements);
const res = await safeFetch(url, {}, {
pay,
policy: "block-ghosts", // block unverified recipients (default)
onCheck: ({ payTo, trust, decision }) =>
console.log(`${payTo}: ${trust.verdict} -> ${decision.allow ? "pay" : "BLOCKED"}`),
});If the recipient is a ghost/sybil, safeFetch throws X402_SAFE_BLOCKED instead of paying.
Policies
block-ghosts(default) — pay only verified, settlement-backed recipients; block unverified.block-unknown— also block when AgentRank can't be reached.warn— pay anyway, but surface the verdict viaonCheck.allow— never block (telemetry only).
Just the check
import { checkAgentTrust } from "x402-safe";
const t = await checkAgentTrust("blockrun.ai"); // wallet or domain
// { verified: true, score: 870, settlement: { usd: 1026, payers: 37, substrate: "x402" }, verdict: "..." }How the verdict works
AgentRank ranks agents by real on-chain settlement weighted by payer reputation (PageRank on the
USDC payment graph, de-noised, sybil-tested). A recipient is verified only if it has actually settled
value that traces back to reputable payers. Sybils and ghosts score ~0 because you cannot fake real
settlement. Method: https://github.com/andysalvo/agentrank.
Also available
- MCP tool — the same check as
check_agent_trustover MCP:https://api.agentrank.info/mcp. - Badge —
https://api.agentrank.info/badge/{wallet|domain}.svg.
MIT. Built by Crest Deployment Systems LLC.
