Micropayments for APIs and AI agents: how pay per request works

What are micropayments for APIs?

Micropayments for APIs are small charges paid for each request a caller makes. A weather lookup might cost a cent, a search a tenth of a cent, a generated file a few cents. The caller pays for exactly what it uses, with no plan, monthly minimum or prepaid balance at the seller.

Pay per request pricing has existed as an idea for a long time. It mostly stayed an idea because the cost of moving a small payment was larger than the payment. AI agents change the demand side: they call many different services, often once each, and they work faster than any person signing up for accounts.

This guide explains why card fees fail micropayments, what developers use instead, and how sub-cent payments work in practice.

Why do card fees fail micropayments?

Card processing charges a percentage plus a fixed fee on every transaction. Stripe's standard US online price is 2.9% plus 30 cents per successful domestic card transaction (Stripe pricing). The fixed 30 cents is the problem. It stays the same whether the charge is $100 or $0.01.

Here is what a seller keeps from a single card charge at that rate:

Below about 30 cents, every card charge loses money. At $1 the fee takes about a third. Card rails work well for larger purchases, which is what they were built for. They were never designed for one call at a time.

Card checkouts also tend to set minimums. On Qatom, guest card checkout has a minimum of about $0.50, so items priced under 50 cents suit agent wallets instead (pick a use case and price).

How do developers charge for small API calls today?

Many API businesses avoid true micropayments for APIs by grouping small charges together. The common patterns are:

• Subscriptions. A monthly plan with a quota of calls.
• Prepaid credits. The caller buys a block of credits by card, and each call draws it down.
• Metered billing. Usage is counted and invoiced at the end of the month.

Each of these spreads one card fee across many calls. Each also needs an account, an API key and a billing relationship with every seller. That works for a developer who uses three APIs every day. It works badly for an agent that needs one call from each of fifty services, once, right now.

What makes AI micropayments different?

AI micropayments have a few needs that human billing does not.

No sign-up step. An agent cannot fill in fifty sign-up forms and store fifty API keys safely. It needs one connection that can pay any seller.

Spending limits set by a person. The agent's owner needs a hard ceiling and an approval step for anything large. A prepaid balance per seller does not give one view of total spend.

A receipt per call. Every call should return a transaction id, so the owner can check what was bought and ask for a refund if needed.

Safe retries. Agents retry failed calls. If a paid call's output is cut off, the purchase may still have gone through, and a retry is a new charge (retry safely).

How do sub-cent payments work?

Sub-cent payments need a rail without a card-style fixed fee on every transaction. Two approaches used for agent payments today:

x402. An open payment protocol created by Coinbase, announced in May 2025 with AWS, Anthropic, Circle and NEAR, and governed since April 2026 by the x402 Foundation under the Linux Foundation. A server answers a request with HTTP 402 "Payment Required", a price and the tokens it accepts. The client signs a payment and retries with an X-PAYMENT header, a facilitator verifies and settles onchain in stablecoins such as USDC on chains such as Base, and the server returns the data (Coinbase). It fits builders already working with stablecoins and public chains. Independent 2026 studies found most raw x402 volume was not real commerce: Visa and Artemis put adjusted volume at roughly $15 million through April 2026, an average of about $0.14 a payment (x402 in 2026).

Wallet-to-wallet settlement on Qatom. Each buyer's AI spends from an agent twin, a small secure wallet that holds USD-TDN, a US-dollar token where 1 USD-TDN equals 1 USD. Payments settle between twins in under one second, with no gas and no public ledger. Qatom charges a 0% take rate and no transaction fees, and the cost does not scale with the value or volume of the payment. The smallest paid price is 0.001 USD-TDN.

These approaches can complement each other. A seller can keep a subscription for heavy users and add pay per request for agents. Our agent payment protocols compared guide covers x402, AP2 and ACP in more depth.

Note that MCP, the open protocol AI assistants use to reach tools, has no payment step of its own (MCP specification). Payment has to come from a layer like the ones above.

How do you set up pay per request on Qatom?

A seller needs one HTTPS endpoint and a catalog item for each price. The full path, step by step, is in the Capabilities Index.

1. Pick one item and a price. Choose something that answers in one call and is cheap to serve. Set a price of at least 0.001 USD-TDN with no more than three decimals, and add a free sibling check (B1).
2. Build the endpoint. Put the product behind one HTTPS endpoint; the Worker template is the fastest start (B2).
3. Guard the paid route. Protect it with a secret Authorization header that only Qatom sends, and answer 404 without it (B3, B12).
4. Write the agent read-me. Generate an llms.txt that tells agents how to call and pay you (B4).
5. Create the catalog item. Enter the name, descriptor, price, method and endpoint, plus an input schema with additionalProperties set to false (B9, B11).
6. Buy it once. Test each item with a real purchase through the master MCP before announcing (B19).
7. Serve each sale once. For anything costly to serve, verify the X-Qatom-Transaction-Id on each call and keep the ids you have served (B20).
8. Announce. Confirm the item shows in catalog_search, then share your llms.txt (B21).

Buyers connect any MCP client to the Qatom master MCP at https://mcp.m.todaq.net/mcp, find your item with catalog_search and pay from their agent twin. Purchases under the buyer's auto-approve limit go through without asking. For the full seller walkthrough, read how to charge AI agents for API calls. The API reference and SDKs are in the developer docs.

How should you price micropayments for APIs?

A few rules keep per-call pricing clear for agents and fair for you.

• One item, one price. Several prices means several catalog items, each with its own description of what the higher tier adds.
• Price from your cost. Start from what one answer costs you to produce, then add your margin.
• Offer a free check. A free sibling item lets an agent confirm the input works before it pays.
• Return complete answers. Each call should stand alone, with a one-sentence summary and clear errors.
• Mind the plan limits. Plans set how many agents, items and transactions the account gets. Builder is free with 1 agent, up to 5 items and 25 transactions a month; Pro is $20 a month and Max $300 a month (plans and pricing).

What does it cost to take micropayments on Qatom?

Qatom takes 0% of each sale and charges no gas and no transaction fees. Revenue lands in USD-TDN, and you can split it across up to 200 payees (split revenue).

To move money to a bank, you cash out to a US or Canadian account. A cash-out needs at least 10 USD-TDN, carries a flat $1 fee and arrives by ACH in 2 to 5 business days (cash out). RBC holds the reserve funds backing USD inside Qatom. The launch offer on the site gives 6 months of Max free, with no card to start. To talk through a larger catalog or a VPC, on-premises or air-gapped deployment, contact us.

Next steps

• Follow the seller steps in how to charge AI agents for API calls.
• Read the basics in what are agentic payments.
• Check Qatom plans and pricing and start on Builder for free.