Since July 13 an AI agent has been able to pay two dollars in USDC on Base and get back a verdict from mcp.djzs.ai. The payment settles wallet-to-wallet with no custodian in the middle. The verdict comes with a hash anyone can recompute and, since September 1, an on-chain attestation naming the agent that paid. The discovery files are all there the payment manifest under .well-known, the llms.txt, the OpenAPI, the agent card. Every piece of plumbing an agent needs to find a service, understand it, pay for it, and prove it did.
Number of agents that have paid, not counting my own wallets: zero.
I could dress that up. I have watched other people dress it up. The honest version is more useful to you, because the reason nobody has paid is not that the plumbing is wrong. I have the logs. Unpaid requests get a correct 402 challenge. Paid requests settle and attest. The discovery surfaces validate against the current spec. The plumbing works. What has not arrived yet is the traffic the marketplace that indexes agent-payable services is slow and inconsistent, and agent-to-agent commerce in September 2026 is real but thin.
So I am sitting on a working, production-tested stack for a market that is early, and I have spent the last year learning every way that stack fails to come together. Which is a strange asset to hold, until you notice who else needs it.
Every API owner who has read a headline about agents paying for services and has no idea how to receive the payment.
That is a lot of people. If you run a SaaS product, a data feed, a prediction-market tool, a research endpoint anything with a URL that returns something useful then an AI agent holding USDC cannot pay you today. It cannot complete your Stripe checkout. It cannot find you unless you publish the files it reads. It cannot prove to its operator that it paid you and got what it paid for. Stripe stays for your humans; the agents need a different door, and almost nobody has built one.
The specs to build that door are public. The reason to hire someone anyway is that the specs describe the pieces, not how they fail to fit. It took me from July to September to get discovery, settlement, and receipts all correct in production at the same time. You are buying those weeks.
Fixed scope, fixed price, two weeks, your keys. Details are at djzs.ai/build; the short version:
Agent-ready API — $3,500. Your existing endpoints wrapped as an MCP server on Cloudflare Workers, an x402 paywall in front of the routes you choose, funds settling directly to a wallet you control, every discovery file an agent reads, and a submission to the marketplace index. Ten working days. Source in your repo. A runbook and a probe script so you can check it yourself every morning.
Receipt layer — $2,500. Every paid response carries a hash of the request and an attestation on Base under your own schema, from an attester wallet you hold. Plus a static verifier page, so a customer, a court, or a DAO can recompute the receipt without trusting you. This is the part that turned out to matter most in my own build, and the part almost nobody else ships.
Business manifest — $400. For a physical business with a price list and a booking page: a signed machine-readable listing under your domain so an agent shopping on a customer's behalf can find you and quote you. No code on your side. Three days.
Keep-alive — $600 a month, optional. The spec moves. Facilitators change. Worker runtimes update. I track it and re-validate your surfaces so they keep passing.
Half up front on a signed scope sheet, half on acceptance. Acceptance is not my opinion it is a list on the scope sheet: an unpaid call returns a valid 402, one paid call settles to your wallet with a transaction hash recorded, the tools list from the production URL, all discovery files return 200 and validate. If those four things are not true on day ten, the second half is not invoiced.
Day zero, a thirty-minute call and a scope sheet we both sign. Day three, the MCP server is on a staging Worker and you are calling it from Claude or Cursor nothing paid yet. Day seven, the paywall is live on staging and the first USDC settles to your wallet, not mine. Day ten, production deploys from your Cloudflare account, discovery files go live, you get the runbook and the probe script.
You set every secret yourself, in your own account, from the runbook. I never hold a wallet key, an API key, or a deploy token past the call where you type it in. That is not a courtesy. If the person who built your payment rail also holds your keys, you did not buy a payment rail; you bought a dependency.
That agents find you on day one.
I have been staring at the indexing problem since August. I build the surfaces correctly, I submit the listing, I hand you the script that checks it every morning and then traffic is a distribution problem we solve after the plumbing exists, the same way it is for me. Anyone who tells you otherwise has not shipped one of these and watched the logs.
I also do not do trading signals, custody, token launches, or chains other than Base without a written reason. If your ask is one of those, I will say so on the first call and point you somewhere else.
The honest answer on now: because a year of building for a market that has not shown up has produced exactly one thing of certain value, and it is the knowledge of how to build for that market. Selling the product to agents comes later. Selling the skill to the humans who own the APIs comes first, and every one of those builds is a live test of whether the product is worth anything because every client I wire up is an API an agent can pay, and a wallet that can pay mine.
The answer on me is the receipts. Two paid tools live on Base mainnet since July. Verdict hashes that have recomputed byte-identical on every re-run since. Attestations on-chain since September 1. A pre-registered outcome study whose first sealed day was anchored to permanent storage this week, at a root you can look up. A ruleset page that publishes the measured band instead of the best run. I hold my own work to that standard in public, and I will hold your build to it too.
Send three lines by DM to @Djzs_ai on X:
What the API does, in one sentence.
Who should be able to call it humans, your own agents, anyone's agents.
When you need it live.
You get a yes-or-no on fit within one working day and a scope sheet within three. Two engagements open at a time.
Plumbing first. Then the traffic.
END_TRANSMISSION. //