How to sell an API, dataset or service to AI assistants

How do you sell an API to AI assistants?

To sell an API to AI assistants, three things have to be true. An agent must be able to find your service from inside a chat, understand what it returns before paying, and pay without a person filling in a card form for every call.

Assistants act through tools, so your product has to show up as a tool with a price, a description and a schema. This guide covers each part, from discovery to announcing a finished store.

The same steps work for an API, a dataset or a service such as a file, a video stream or a report. On Qatom the path follows one value chain: set up, publish, distribute, transact, settle, report. Every step is also in the Capabilities Index.

How do AI assistants find tools through MCP?

Many assistants reach outside tools through MCP, the Model Context Protocol. It is an open protocol for connecting AI assistants to tools and data, and remote servers authorize with OAuth 2.1. MCP itself has no payment step, which is the gap a storefront fills.

On Qatom, a buyer adds one connection, the master MCP at https://mcp.m.todaq.net/mcp, in Claude, Grok, Muse or any other MCP client (A1). That one connection gives the assistant every public catalog item plus wallet tools.

When a user asks for something, the assistant calls catalog_search with plain words such as "weather" (A10). Each result carries an id, name, description, price, seller and input schema. The assistant then reads the listing, checks the price and buys with agent_checkout, which pays from the agent's own wallet and returns your answer in the same call.

That flow tells you where to spend your effort. Your name and description help an agent match you to a query. Your description and schema help it decide whether to buy. Your endpoint's answers shape whether it buys again.

What is an agentic storefront?

An agentic storefront is a set of items that AI agents discover, order from and pay for on their own, inside limits a person sets. On Qatom, each catalog item is one thing for sale at one price with one endpoint (B9). The item name becomes the tool name agents see.

Payment, delivery and a verifiable receipt happen in one call. The agent pays in USD-TDN, a US-dollar token (1 USD-TDN = 1 USD), and Qatom calls your endpoint after the payment settles. Settlement takes under one second, and every purchase returns a transaction id and an emailed receipt.

For the payment side in more depth, read how to charge AI agents per API call.

How do you get your API discovered by AI agents?

Follow these steps in order. Each one links to the matching step in the Capabilities Index.

1. Pick one first item. Choose something that answers in one call and is cheap to serve. The seller guide skill helps: npx skills add StarwaterHeaven/qatom-seller-guide --skill qatom-seller-guide (B1).
2. Register a seller account. Sign up at dashboard.qatom.ai, choose a plan and start your MCP server (B34).
3. Build one HTTPS endpoint. Start from the Worker template, which reads one catalog.json, validates inputs, guards the paid route and serves /llms.txt (B2).
4. Guard the paid route. Check a secret Authorization header that only Qatom sends, and answer 404 when it is missing or wrong (B3, B12).
5. Write llms.txt. Generate it from your catalog and serve it at your site root (B4).
6. Create the catalog items. Set name, descriptor, price, method, endpoint, description and input schema (B9 to B11).
7. Buy each item once. Test through the master MCP before you tell anyone (B19).
8. Check search, then announce. Confirm each item in catalog_search, then share your read-me and add your store to STORES.md (B21).

What should an agent read-me (llms.txt) include?

llms.txt is a Markdown read-me for language models, served at your site root. Agents that find your items look for it before calling.

Generate it from catalog.json or from the live public catalog (GET https://pay.m.todaq.net/v4/catalog/items). Never write it by hand, and regenerate it on every catalog change so it never drifts from what is listed. Include:

• Your seller name, a short summary and a Generated timestamp.
• How to call: catalog_search, then agent_checkout.
• Worked examples and a table of items with prices.
• Input schemas, every error code with its next step, and your free feedback item.

Put the llms.txt URL in every item description. The Qatom Demo store serves its own at https://qatom-demo.agentkepler.tech/llms.txt, listing each item, its price and its inputs.

How do you write item names and descriptions agents will choose?

The name is the tool name agents see, so make it predictable. Use a shared prefix in the form "Brand: item", such as "Qatom Demo: space station video". A shared prefix lets buyers, and you, search for your whole store at once.

The descriptor is a separate field: it is the line on the buyer's receipt and card statement. Renaming an item later does not change the descriptor, so set it with care.

The description is what an agent reads to decide whether to buy (B10). Keep it at or under 1,024 characters.

One catalog item has one price. If you want several prices, create several items. Prices go down to 0.001 USD-TDN per call, with no more than three decimals, and 0 makes an item free.

What makes an agent choose your item and repeat it correctly?

The AAIF Ambassador talk "Agentic commerce for sellers" at MCP Dev Summit Toronto 2026 argued that agents compete on the payload, so every fact an agent must repeat (fees, currency, times) belongs in typed fields. Its rule: "If it isn't in the response, it doesn't exist."

• Put every fact the buyer must know (price, currency, limits, delivery time) in the item description and in typed fields of your response, not only in prose.
• Name the item by the outcome it delivers, so the agent can match it to the user's request.
• Return errors that say what to do next, so the agent can recover without guessing.

What should the input schema look like?

The input schema tells agents exactly which arguments to send (B11). Qatom forwards only declared arguments and rejects the rest, so a loose schema leads to failed calls.

Read the schema off your endpoint. Use one required string that identifies the item, and one optional property per real input. Give every property a description with its unit, range and default, and set "additionalProperties": false.

Describe formats instead of giving example IDs or item names. For instance, the Qatom Demo video item takes one optional input, ttl_seconds, from 60 to 86400, default 3600. Never accept raw SQL, code or query strings as inputs; turn them into named inputs.

Return every error with a code, a plain message and a next step, such as bad_value with a 400 status, so an agent can correct itself.

Why add a free feedback item?

A free item lowers the cost of trying your store to zero. Agents and people can check that your service works, and you collect feedback from real buyers. The seller guide pairs each first item with a free sibling check for the same reason.

The Qatom Demo store sells a video at 0.01 and an agent skill at 0.005, beside a free feedback item. Free items also cost nothing to test.

How do you test before you launch?

A real purchase is the only full test of price, schema, header and endpoint together (B19). Call your endpoint directly first, because Qatom may relay a failed call as a generic payment error.

Then connect the master MCP, put a small amount in your agent wallet and buy each item with agent_checkout: once alone, once with inputs and once with a bad input. Check that the receipt shows your endpoint's answer and a transaction id.

A call your endpoint rejects can still be charged, and a retry is a new charge. For anything expensive to serve, such as downloads or stream keys, read the X-Qatom-Transaction-Id header on each call and look it up, so each sale is served once (B20).

Can I list my service in ChatGPT and other assistants?

ChatGPT is among the assistants and channels listed on qatom.ai, along with Claude, Microsoft Copilot, Gemini, Muse, WhatsApp, iMessage, Grok and Hermes Agent. Buyers in any MCP client can add the master MCP and reach your public items. Treat these as places your storefront can be found, and check the steps for each client in A1.

Other routes into assistants exist. The Agentic Commerce Protocol, co-developed by Stripe and OpenAI, powers Instant Checkout in ChatGPT and passes the merchant a scoped payment token. It rides card rails and suits physical goods at checkout. At Stripe's standard 2.9% + 30¢, a $0.10 charge would cost about $0.30 in fees, so per-call pricing works better on a wallet that settles without card fees. These approaches can complement each other. See agent payment protocols compared for the wider picture.

How do you announce your store?

Announce only once your items show in catalog_search and pass your launch checks. Search the master MCP for your brand prefix and check each public item's name, price, seller name and description. Delist or archive test and placeholder items first.

Then tell people where to find you:

• Share your llms.txt URL wherever developers and buyers look.
• Add your store to STORES.md.
• If you run your own MCP server, consider publishing it to the official MCP Registry, which exists for publishing servers.

Private items never show in catalog_search; they are reachable only through your own MCP instance. Choose public or private when you create an item, because the choice is fixed (B14).

How much does it cost to sell to AI assistants?

Qatom takes 0% of each sale, charges no transaction fees and has no gas. Cost does not scale with the value or volume of the payment.

The site's launch offer gives 6 months of Max free, with no card to start. Cashing out to a US or Canadian bank needs at least 10 USD-TDN, costs a flat $1 and takes 2 to 5 days by ACH (C26). See pricing for current plans, and the developer docs for the API and SDKs.

Next steps

• Follow the seller steps from B1 in the Capabilities Index.
• Read how to monetize an MCP server if you already run one.
• Talk to the team about hosted, VPC, on-premises or air-gapped setups on the contact page.