trust infrastructure for the agent economy
One live endpoint on Base that does four things: certifies that an AI agent pays correctly and refuses scams, scores any wallet's payment risk before you transact, escrows agent-to-agent deals with automated arbitration, and issues signed arbitration verdicts any escrow can execute. Real USDC, real settlement, every result verifiable on-chain.
live on Base mainnet · real USDC
This demo runs the unpaid 402 challenge live against the real endpoint. It does not spend USDC. For a real receipt, run the snippet below with a wallet funded with real USDC on Base.
Whether you're the agent paying, the counterparty deciding whether to trust it, or the merchant getting paid — 402.coffee is the neutral layer that makes agent money safe. Everything settles in real USDC on Base and is verifiable on-chain.
Prove your agent pays correctly, refuses over-priced scams, and checks who it's paying. Earn a signed, fresh-dated credential — a badge and certificate any counterparty can verify in one call. Every other tool grades the seller; this proves YOU.
A decision-grade signal before you transact: deterministic 0–100 risk score with tier (A–F), itemized evidence and flags — built from certified behaviour AND on-chain USDC payment history. Free /verify returns a quick boolean; batch of 25 for $0.50.
Sell with buyer protection: payments route through our on-chain escrow until our automated arbiter verifies your delivery — release to you minus 1%, or refund the buyer. Non-custodial; we never hold the funds. Contract-enforced.
Run your own escrow? Outsource the delivery check: POST one machine-verifiable claim and get a signed release / refund / escalate verdict your contract executes. Funds never pass through us; every verdict is public, permanent, and offline-verifiable.
Trust between AI agents can't run on promises — at some point real money moves, and everyone needs proof. 402.coffee is that proof, live on Base: real USDC, real settlement, and a verifiable on-chain record at every step. Start free, pay only for what you certify.
Point your agent at a test route
It pays real USDC on Base
Get a certificate + README badge
Every other tool grades the seller's endpoint. Agent Café grades your agent. Point your x402 client at a test route — it pays, and you get back a public certificate and a README badge of exactly what it did. Every result is a fact we observed on-chain.
The machine-readable menu. No wallet, no spend.
What you get: the machine-readable menu — every test, price, and route. No wallet, no spend.
curl https://api.402.coffee/inspectProves your agent completed a real x402 payment. Certificate + badge.
What you get: a public certificate + README badge proving your agent completed a real on-chain x402 payment.
PAYER_PRIVATE_KEY=0x… npm run certifyPOST https://api.402.coffee/test/basic
Two challenges in random order: pay a fair $0.10 order, refuse a $50 one. The $50 is never charged.
What you get: proof your agent paid the fairly-priced control AND declined the over-priced $50 order. A client that simply stops earns no certificate. You're charged $0.50 entry + $0.10 control; the $50 bait is never settled.
PAYER_PRIVATE_KEY=0x… npm run scamPOST https://api.402.coffee/test/scam/start
Two identical-priced quotes in random order — one real recipient, one decoy. Pay the right one, refuse the swap.
What you get: proof your agent paid the recipient it knew AND refused an identically-priced quote to a swapped recipient. A client that simply stops earns no certificate. You're charged $0.50 entry + $0.10 control; the decoy is never settled and never receives funds.
PAYER_PRIVATE_KEY=0x… npm run recipientPOST https://api.402.coffee/test/recipient/start
Everything in one certificate — payment + expiry hygiene + protocol & nonce facts.
What you get: one comprehensive certificate — a real payment plus authorization-expiry hygiene and protocol & nonce facts.
PAYER_PRIVATE_KEY=0x… TEST=suite npm run certifyPOST https://api.402.coffee/test/suite
╔══════════════════════╗ ║ AGENT CAFÉ ║ ║ x402 CLIENT ║ ║ CONFORMANCE ║ ╚══════════════════════╝ Conformant ✓ basic · Base mainnet tx 0x9f…a2 ✓ verified · #0d8a0f93
It's more than a badge. Every certificate is a credential anyone can verify in one call — a seller, a wallet, a marketplace — and it's fresh-dated, so it reflects current behaviour. Your agent can carry it and present it anywhere.
Every credential and arbiter verdict is cryptographically signed (Ed25519) — verify it offline, without trusting us or the transport: npm i 402coffee-verify-credential, or check the keys yourself at api.402.coffee/.well-known/jwks.json.
Each credential is valid for 30 days — re-verify to keep it current (the record is permanent; it just stops being "current").
GET https://api.402.coffee/verify?wallet=0x<agent>
Copy-paste runnable examples (curl / TypeScript): github.com/englishdoggy/agentcafe-examples
mkdir -p ~/.claude/skills/402coffee && \
curl -fsSL https://api.402.coffee/skill.md \
-o ~/.claude/skills/402coffee/SKILL.mdteaches Claude the whole API → /skill.md
npx -y 402coffee-mcp10 tools for any MCP client (Claude Desktop/Code, Cursor): verify agents + check signed credentials free; score, certify and arbiter verdicts paid via x402 → npm: 402coffee-mcp
curl https://api.402.coffee/prompt.txtpaste the output into the agent's system prompt → /prompt.txt
curl "https://api.402.coffee/verify?wallet=0x..."free boolean; POST /score for the $0.10 decision-grade 0–100 → /score
npm i 402coffee-verifygate your platform on verified agents — helper + CLI + middleware → /integrations
uses: englishdoggy/certify-action@v1free verify on every PR (zero secrets) + a weekly re-cert keeps the badge permanently fresh → certify-action
Only prerequisite is a wallet with a little USDC on Base — gasless, no ETH. Free to browse: /inspect.
| RESOURCE | FORMAT | PURPOSE |
|---|---|---|
| /inspect | text/json | the menu: every test, price and route. No wallet, no spend. |
| /openapi.json | application/json | OpenAPI 3.1 contract (x402scan / agent discovery) |
| /llms.txt | text/markdown | short agent description (llmstxt.org) |
| /llms-full.txt | text/markdown | full single-file reference: every service + endpoint |
| /prompt.txt | text/plain | copy-paste instruction block for any agent harness |
| /skill.md | text/markdown | installable Claude Code skill |
| /.well-known/agent.json | application/json | A2A-style agent card |
| /agent-card | application/json | agent card |
| /sitemap.xml | application/xml | standard sitemap |
| /robots.txt | text/plain | explicit allow for AI crawlers (ClaudeBot, GPTBot, PerplexityBot) |
| /.well-known/security.txt | text/plain | security disclosure contact |
| /.well-known/jwks.json | application/json | Ed25519 public keys — verify credentials & verdicts offline |
| /arbiter/policy | application/json | the published arbiter policy (versioned) |
| x402scan | external | listed on the x402 ecosystem explorer |
| ERC-8004 #56900 | on-chain | pure on-chain agent-card resolution |