How to sell an API to AI agents
The short version, with links to the deep dives. Written by a seller with real, small, verifiable revenue — not a projection.
"AI agents" are starting to hold their own budgets and make their own purchase decisions mid-task — a research agent paying for a data lookup, a trading bot paying for a market-regime read, a pipeline paying for a specific fact it doesn't have. If you have an API and want machines, not just humans, as customers, there are two things you actually need: a payment rail a machine can use without a human in the loop, and something worth paying for. This page is the map; the three pages below are the depth.
The two rails that exist today
x402 is the agent-native one: your endpoint returns HTTP 402 with a price, the caller's wallet pays USDC on Base, the request retries with proof of payment attached. No signup, no API key, settlement in seconds. We run this ourselves — live at api.stelardigital.com, 18 settled calls, $0.235 lifetime revenue so far, published honestly at /open.
Traditional API marketplaces (RapidAPI and similar) still work, but they're built for a human developer to sign up with a credit card, not for an autonomous agent to transact mid-task. We run one of those too, in parallel, and compare the real results at x402 vs RapidAPI.
There's also a discovery layer worth knowing about: MCP (Model Context Protocol) servers expose your API as tools an agent framework can call directly, and can sit in front of either payment rail. We run an MCP server in front of our x402 endpoints for exactly this reason.
What we've learned selling on these rails ourselves
- What sells: proprietary data exhaust from a system you already run — not a wrapper around something an agent's own tools already do. We retired seven commodity-wrapper endpoints (summarize, extract, fact-check style utilities) after they produced zero of our 18 settled calls. Full breakdown: x402 for indie builders.
- How to price it: in fractions of a cent for a single fact, not dollars for a bundle. Our own $1 and $2.50 bundles have settled zero calls; our $0.005–$0.05 raw-data endpoints account for all 18. Full data: x402 pricing.
- Which rail to pick: depends on whether your buyer is a human with a credit card or a machine with a wallet. We run both and report the real numbers side by side: x402 vs RapidAPI.
Who this is actually for
If you already run something real — a classifier, a scraper with a unique corpus, a bot that produces signal nobody else has, a pipeline that turns raw data into a specific useful shape — and you've never thought of it as a sellable product because there's no obvious human customer for it, this is the gap. Agents are starting to be that customer. They don't need a landing page, a pricing tier comparison, or a sales conversation; they need a machine-readable price and a way to pay it without asking a human for a credit card. That's a much lower bar to clear than building a SaaS product, which is exactly why it's worth doing before the easy claims on x402-practitioner search terms are gone.
It's also worth being honest about what this isn't: it's not a way to monetize a thin wrapper around a public API or a summarization prompt. We tried that with seven endpoints and none of them sold. It's a way to monetize something you already had to build for yourself anyway, at a marginal cost close to zero, to a buyer type that didn't exist as a meaningful market two years ago and is growing now.
The shortest possible starting checklist
- Find data or logic you already run for yourself that nobody else has access to.
- Wrap it with an x402 payment middleware and price it in cents.
- Publish a machine-readable catalog (
/.well-known/x402or similar) so discovery crawlers and agent frameworks can find it without a human reading a landing page. - Emit verifiable receipts so a buyer can prove what they paid for and got — we maintain an open-source format for this: x402-receipts (MIT).
- List it in the discovery catalogs that exist (Coinbase's CDP x402 catalog, MCP registries, x402 Bazaar) and watch settlement, not traffic.
None of this is theoretical. Our own numbers — the good, the small, and the zero — are public and updated live at /open, and we re-run a full census of the x402 economy monthly at /state-of-x402. If you're deciding whether this is worth building, read those before you read anyone's pitch deck, including ours.
We run this business on these rails ourselves — live endpoints, open-source receipts, honest numbers.
See the live x402 API → x402-receipts on GitHub Read our real revenue at /open