Agent Deals

Find the cheapest working paid API for a task, and check any x402 payment before paying (GO / CAUTION / STOP on price, service health and wallet safety). Remote MCP, free checks.

Gehosteter MCP-Server

npx add-mcp 'https://agent-deals.gm-tools.workers.dev/mcp'

Installiert in Claude Code, Codex, Cursor und mehr

Dokumentation

Safe, fairly priced tools for AI agents

Pay-per-use tools that agents can call with x402 (USDC on Base). No signup, no API key. Prices shown upfront, and you're never charged when a call fails.

The tools

Status comes from our own monitor, which checks every listed service the way an outside agent would.

Pay Safe + Agent Deals

working · checked 2 days ago

Before an agent pays any x402 service: is it working, is the price fair, is the wallet safe? GO / CAUTION / STOP. Plus the cheapest working service for any task, from ~25,000 tracked.

Pay Safe and service checks: free. Cheapest-service search: $0.002 (20 free a day).

agent-deals.gm-tools.workers.dev · docs for agents

Trust Check

working · checked 18 h ago

Is this token safe to buy? Is this address safe to send to? Real buy-and-sell simulation, owner powers, scam lists and address-poisoning detection on Base.

$0.003 per token check, $0.002 per address check (5 free a day each).

trust-check.gm-tools.workers.dev · docs for agents

Easy Reader AI

working · checked 18 h ago

Any public web page or PDF as clean Markdown, with title, author, date and links. Or just the page's metadata.

$0.003 per page, $0.002 for metadata.

agent-reader.gm-tools.workers.dev · docs for agents

Proof: how accurate is Trust Check?

Tested 2026-09-28 on real Base tokens, against answers that are certain: famous tokens; established tokens that many different wallets really sold in the last day; rug pulls proven by the blockchain's own records; and likely honeypots (people bought, nobody sold, and an independent sell simulation fails). We didn't use another checker's opinion as the answer key.

0 of 19 famous tokens (USDC, cbBTC, AERO…) wrongly called scam or high risk

0 of 87 established tokens (7+ days old, sold by 20+ wallets in the last day) wrongly called unsellable

19 of 19 confirmed rug pulls (the blockchain shows all liquidity removed after trading) flagged as don't-buy

Some established tokens still get high_risk or scam for other reasons (one wallet holding a large share, or a scam-list flag). Those are risk warnings, not claims that the token can't be sold.

What this test made us fix:

  • 2026-09-28: this test found that Trust Check described drained pools (rug pulls) as a 100% tax. It now reports them as an empty pool / likely rug pull. The rug-pull score above is after that fix.
  • 2026-09-28: an unlocked-liquidity warning on long-established tokens (e.g. DAI, based on a small side pool) is now a caution instead of high risk.

Full results, token by token, with the evidence for each answer: benchmark.json. Automated checks can't catch every scam, so keep spending limits in place.

Live numbers

606 calls answered in the last 30 days (8 paid, 598 free)

28,860 paid agent APIs monitored, from 3,013 sellers

1,482 services we caught charging more than their public listing

Our promises

  • Prices upfront. Every price is in the 402 answer before anything is paid. No hidden fees, and we never take a cut of what agents pay other services.
  • Never charged for errors. Failed calls don't settle a payment.
  • We never hold your money. Payments go straight from your wallet to ours: 0x9a62B1fc2337A9437038D97105F75c369505b563. Pay Safe checks that the wallet in any payment request matches this one.
  • Rankings can't be bought. Agent Deals ranks services by working status, price, uptime and speed only.
  • Accuracy is published, including our misses, and re-tested as the checks change.
  • Minimal data. We keep daily counters and a salted hash of callers, not the content of your requests.

Tell us what's wrong, or what you'd like

Found a wrong answer, a bug, or have an idea? It takes 10 seconds, and every report is read. Agents can POST the same to /v1/feedback.