@agentradar/x402-trust

A pre-pay trust gate for x402 agent payments. Before your agent settles an HTTP 402, score the payee wallet with AgentRadar and block or warn when the recipient is a known scam or low-trust wallet.

Think of it as the "Stripe Radar" for agent payments — a fraud check at the one universal chokepoint every paying agent passes through.

• Library-agnostic. Works with any x402 client (@x402/fetch, x402-axios, custom). It keys off the x402 protocol's 402 response (accepts[].payTo), not a specific client's internals.
• Zero runtime dependencies. Uses the global fetch (Node 20+).
• Fail-open or fail-closed. "block" mode throws on a bad payee; "warn" mode logs and proceeds.

Powered by AgentRadar's on-chain trust scoring (ERC-8004 reputation + static analysis + scam-DB). Verdicts: TRUSTED · VERIFIED · CAUTION · RISKY · BLOCKED.

Install

npm install @agentradar/x402-trust

Quick start — native x402 lifecycle hook (recommended, v0.2.0+)

x402 clients expose an official lifecycle-hooks API. createTrustHook plugs straight into x402Client.onBeforePaymentCreation, so the trust check runs on every transport (HTTP, MCP) right before the payment payload is created — and blocks via the protocol's own { abort, reason } mechanism:

import { x402Client } from "@x402/core";
import { createTrustHook } from "@agentradar/x402-trust";

const client = new x402Client();

client.onBeforePaymentCreation(
  createTrustHook({
    policy: { mode: "block", minScore: 40 }, // block scams + anything under 40/100
    onDecision: (d) =>
      console.log(`[x402-trust] ${d.result.address} → ${d.result.verdict} (${d.result.score})`),
  }),
);

Failure semantics: if the AgentRadar API is unreachable, "block" mode aborts the payment (fail-closed) and "warn" mode proceeds (fail-open).

Alternative — wrap your paying fetch

For clients that don't use @x402/core (or pre-v2 setups), the original fetch wrapper still works:

import { wrapFetchWithPayment } from "@x402/fetch";
import { wrapFetchWithTrust } from "@agentradar/x402-trust";

// 1. Your normal x402 paying fetch
const payFetch = wrapFetchWithPayment(fetch, walletClient);

// 2. Gate it: AgentRadar scores the payee before any USDC moves
const trustedFetch = wrapFetchWithTrust(payFetch, {
  policy: { mode: "block", minScore: 40 }, // block scams + anything under 40/100
  onDecision: (d) =>
    console.log(`[x402-trust] ${d.result.address} → ${d.result.verdict} (${d.result.score})`),
});

// 3. Use it exactly like fetch. Blocked payees throw TrustBlockedError before payment.
const res = await trustedFetch("https://some-x402-service.example/api");

Standalone checks

import { checkTrust, decide } from "@agentradar/x402-trust";

const result = await checkTrust("0xPayeeAddress");
// { address, score, verdict, riskFlags }

const decision = decide(result, { minScore: 50 });
if (!decision.allowed) throw new Error(decision.reason);

Gate inside your own 402 handler (e.g. axios)

If you handle the 402 yourself, pass the response body to assertPayeesTrusted:

import { assertPayeesTrusted } from "@agentradar/x402-trust";

// `body` is the parsed 402 response ({ accepts: [{ payTo, ... }] })
await assertPayeesTrusted(body, { policy: { mode: "block" } }); // throws on a bad payee
// ...then run x402-axios' payment retry

API

| Export | Description | |---|---| | createTrustHook(opts?) | v0.2.0 — Returns a native onBeforePaymentCreation lifecycle hook for x402Client (@x402/core). Scores the selected payee and aborts via { abort, reason } per policy. | | wrapFetchWithTrust(payFetch, opts?) | Returns a fetch that probes for 402, scores each payee, enforces policy, then delegates to payFetch. | | checkTrust(address, opts?) | Calls GET {baseUrl}/verify?target=…; returns { address, score, verdict, riskFlags }. Cached in-memory (TTL configurable). | | decide(result, policy?) | Applies a TrustPolicy → { allowed, reason, result }. | | assertPayeesTrusted(payload, opts?) | Extracts accepts[].payTo from a 402 body and enforces policy (throws in block mode). | | extractPayTo(payload) | Pulls unique, lower-cased payee addresses from a 402 body. | | createTrustGate(opts) | Factory binding one option set to all helpers. | | TrustBlockedError | Thrown in block mode; carries .decision. |

TrustGateOptions

| Field | Default | Notes | |---|---|---| | baseUrl | https://api.vvpro.ai | AgentRadar API base. | | apiKey | — | Sent as x-api-key (raises rate limits). | | policy.mode | "block" | "block" throws on a bad payee; "warn" continues. | | policy.blockVerdicts | ["BLOCKED"] | Verdicts that fail the gate. | | policy.minScore | — | Block below this composite score (0–100). | | cacheTtlMs | 60000 | In-memory cache for /verify. 0 disables. | | fetchImpl | global fetch | Override for tests / older runtimes. | | onDecision | — | Callback per payee decision (telemetry/logging). |

How it works

agent ──probe──▶ x402 service           (no payment)
        ◀─402──  { accepts:[{ payTo }] }
agent ──verify─▶ AgentRadar /verify      (score each payTo)
        ◀──────  { score, verdict }
  block? ──yes──▶ throw TrustBlockedError (no funds move)
         ──no───▶ payFetch() settles the 402 normally

Develop

npm install
npm test        # vitest
npm run build   # tsc → dist/

MIT © AgentRadar