Tender

Tender is a buyer-side comparison shopper for paid agent services on OKX.AI.

Give Tender several ASP endpoints that claim to do the same job. It makes one real, x402-settled purchase from each, measures what actually came back: price paid, response latency, delivery completeness, and returns a ranked scorecard that names the best value, with an on-chain transaction as proof for every purchase.

The problem: once agents pay per call, they face a question nobody answers: of the services that all claim to do this, which is actually the best value right now? Buying the first one you find means overpaying, waiting on slow endpoints, or paying for thin results. A service can even take payment and deliver nothing.

What Tender does differently: where scanners and guards check whether a single service is safe, Tender compares services by actually shopping. It test-buys from each competitor and ranks them on a transparent value score (price 45%, latency 25%, delivery 30%), explaining in plain language why each placed where it did. Services that take payment but deliver nothing are caught and excluded: a broken service can never win on price alone.

How it works: one MCP tool call: compare_services(targets, paid_tool) runs the whole shopping trip: probe each candidate's reachability and x402 paywall, make a real settled purchase from each, then rank and return an evidence-backed verdict at a shareable permalink.

Provably real: every verdict is backed by an on-chain settlement, not a simulation. See a live comparison with real transaction hashes at the demo link.

Let your agent buy the best, not the first.

Built Tender end to end during the hackathon as a working A2MCP service on OKX.AI:

• Designed and implemented the comparison engine: probes each candidate ASP, makes one real x402-settled purchase from each, and ranks them on a transparent value score (price, latency, delivery).

• Built real on-chain settlement from the OKX x402 SDK up, a direct EIP-3009 transferWithAuthorization settlement path with a bounded, capped payer wallet. Every comparison produces genuine, explorer-verifiable transactions.

• Implemented plain-language verdicts that explain why each service ranked where it did (e.g. "delivers 14% more data but costs 3× more and runs slower than the winner"), with strict honesty. The winner only claims an axis it actually leads.

• Added a full reliability floor: SSRF guards, per-call timeouts, spend caps, a kill switch, and sequential on-chain settlement to avoid nonce collisions across multiple purchases.

• Built the A2MCP surface (compare_services), server-rendered comparison permalinks, a landing page, and a comparable-vendor test harness for demonstrating value tradeoffs.

• 26 passing unit + end-to-end tests. Deployed on public infrastructure with real on-chain settlement confirmed on-chain.

Not fundraising. Built as a bootstrapped MVP with a clear path beyond the hackathon: billing-flow and value comparison for the broader paid-agent / x402 economy.