# How to make your API discoverable to AI agents An agent finds a paid API through a facilitator catalogue, by probing the host directly, through the settlement record of payments you have received, or through an index that joins all three. Being findable means registering with catalogues, publishing machine-readable files at well-known paths, and accepting payments that leave a public trace. Published: 2026-10-14 Last reviewed: 2026-10-14 Publisher: Roundhouse AI Ltd. Source: https://roundhouseai.io/resources/make-your-api-discoverable-to-ai-agents --- ## The four routes, and why you need more than one An agent that has never heard of you finds you one of four ways. Each has a failure mode, which is why relying on one leaves most of your potential buyers unable to reach you. | Route | How it works | Misses | | --- | --- | --- | | Facilitator catalogue | You register; the catalogue publishes you | Anyone who never registered | | Asking the host | The agent fetches well-known paths on your domain | Anyone who does not know your domain | | The settlement record | Your receiving address appears in payments | Anyone who has not been paid yet | | A joined index | All of the above, reconciled | Whatever all of them miss | The second column is the work. The third is why the work has to be done more than once. ## 1. Register with catalogues The highest-leverage single action, and most sellers stop after one. Several facilitators and directories publish resource catalogues that agents and indexers crawl. Registering with the one your facilitator runs is obvious. Registering with the others is what separates a seller that is findable from one that is findable by Coinbase's buyers. What to get right when you register: - **The payTo address must match your live 402.** Catalogue entries and live terms drift, and the live 402 is what a buyer actually pays. A mismatch means a buyer that trusted the catalogue pays the wrong address or fails. - **The price convention.** Catalogues usually take dollars; the x402 `accepts` array takes smallest units. Confusing them is a factor-of-a-million error. [The accepts schema](https://roundhouseai.io/resources/x402-payment-requirements-accepts-schema) covers it. - **A real description.** A catalogue entry is often the only prose an agent reads before deciding. "API" is not a description. ## 2. What does a host get asked for? An agent that knows your domain will probe it. What it looks for, in rough order of how often it is checked: - **`/.well-known/x402`** or your facilitator's equivalent, declaring your priced endpoints. - **`/llms.txt`**, a plain markdown index of what your site offers and where. [llms.txt explained](https://roundhouseai.io/resources/llms-txt-explained) covers the format and its limits. - **`/.well-known/agent-card.json`** if you present as an agent rather than only as an API. Note the filename changed during A2A's development and older documentation points at a path that no longer resolves. - **An OpenAPI document** at a predictable path. This is the single most useful thing for an agent deciding whether it can use you, because it answers "what can I call" in a form that needs no parsing of prose. > [!IMPORTANT] > The HTTP status decides, not the content type. Some hosts answer 404 with `content-type: application/json`, and a crawler that checks the content type rather than the status reads a missing document as a present one. If you are building the crawling side, status first, always. ## 3. Make your 402 self-describing This is the route people forget, and it is free. A 402 response that carries full terms is a price list a buyer has never seen before can act on. An agent can probe your endpoint, read the terms, decide, and pay, with no catalogue and no documentation anywhere in the loop. So return real terms rather than a bare 402, keep a free path open so a buyer can see your interface before paying, and make sure your `resource` field identifies what is being bought. [HTTP 402 explained](https://roundhouseai.io/resources/http-402-payment-required-explained) covers the response shape. Keeping a free discovery path open is worth stating separately. A buyer that cannot read your schema without paying will usually not pay. ## 4. Why do payments make you findable? Every x402 settlement is public, so your receiving address accumulates a history whether you plan for it or not. That history is what a buyer screening you reads: how many distinct counterparties have paid you, over how long, at what amounts. It is the one credibility signal you cannot manufacture cheaply and the one a new buyer values most, because it does not require trusting you. [Auditing a service's traffic](https://roundhouseai.io/resources/how-to-audit-an-x402-services-traffic) is the buyer's side of the same record. The practical implication is a reason to choose a well-indexed chain. A settlement nobody indexes is public and invisible at the same time. ## What about ordinary SEO? It still matters, and increasingly the audience is an assistant summarising your page rather than a person reading it. The things that help are unglamorous: answer the question in the first paragraph rather than after three paragraphs of context, use headings that are the questions people actually ask, keep paragraphs short enough to quote, and put facts in tables. Serve a plain markdown version of each page at the same URL plus `.md` if you can, which removes the parsing step entirely. What does not help is serving something different to a crawler than to a person. Beyond the obvious risk, it means the thing being quoted about you is not the thing you can stand behind. ## Frequently asked questions ### How long until an agent finds me? Catalogue registration can be hours to days depending on crawl cadence. Host probing only reaches you if something already knows your domain. The settlement record requires that you have been paid at least once, which is why the first few payments matter disproportionately. ### Do I need to be in every catalogue? Not every one, but more than one. Catalogue coverage is uneven and partial: measurements in 2026 found large sets of live sellers absent from the biggest directories entirely. ### Is an llms.txt file enough on its own? No. It helps an agent that has already reached your domain. It does nothing for one that has not heard of you, which is most of them. ### Should I publish prices publicly? Yes, in your 402 and in your catalogue entries. A price an agent cannot read before committing is a price it will skip in favour of one it can. ### What if I do not use x402 at all? The first two routes still apply: catalogues and well-known files. You lose the settlement-record route entirely, which is the one that produces verifiable credibility rather than claims. ## Where to check this yourself [/services](https://roundhouseai.io/services) is a catalogue built from several of these routes at once, so you can see whether your endpoint appears and what terms are recorded against it. If it is missing or the terms are wrong, that is the same view a buyer gets. [/wallets](https://roundhouseai.io/wallets) shows the settlement history behind any receiving address, which is the credibility half. --- Roundhouse indexes x402 settlements and ERC-8004 identities into one public record. - Every article: https://roundhouseai.io/resources.md - Every surface on this site: https://roundhouseai.io/llms.txt - The index, endpoints and schema: https://api.roundhouseai.io/v0 - A key of your own, $0.01 over x402, no account: https://api.roundhouseai.io/v0/test/x402 - Limits and the SQL surface: https://roundhouseai.io/docs/api