← Articles

How Merchants Accept Payments From AI Agents

Published November 10, 2026Updated November 10, 2026Merchants

Key takeaways

  • Most merchants block agent traffic today because an unknown agent looks the same as a scraper or a fraud bot, not because they do not want the sale.
  • To accept a payment from an agent safely, a merchant needs three things: a verifiable agent identity, proof that a human approved the specific purchase, and a payment that arrives on rails the merchant already uses.
  • Agent identity standards sign each request cryptographically. Examples include Web Bot Auth and Visa's Trusted Agent Protocol.
  • Card networks and protocol groups now define how agent purchases are authorized: Visa Intelligent Commerce, Mastercard Agent Pay, the Agentic Commerce Protocol from OpenAI and Stripe, the Universal Commerce Protocol from Google and Shopify, and Google's AP2.
  • A merchant with a website can usually keep its existing checkout and processor. A merchant with an MCP server can price its tools and run paid work only after the user approves the amount.

Merchants can accept payments from AI agents when three conditions are true: the merchant can verify which agent is calling, the merchant can see that a human approved the specific purchase, and the payment arrives through rails the merchant already trusts. Card networks, protocol groups and infrastructure providers now publish standards for each of the three, so a merchant does not have to invent its own.

Why do merchants block AI agents today?

Merchants block AI agents because they cannot tell a legitimate buying agent from abusive automation. Bot defenses, CAPTCHAs and fraud models were built for a world where a human is at the keyboard. An agent that arrives with no identity gets scored as a bot, and the correct engineering decision for an unknown bot is to block it.

  • The agent is anonymous: Without a verifiable identity, the merchant sees automated traffic from an unknown source. It cannot separate a shopping assistant from a scraper or a card-testing script.
  • The human is invisible: Even when the agent is recognized, the merchant cannot see who asked for the purchase or whether that person agreed to it.
  • The checkout expects a person: Sign-in walls, verification prompts and one-time codes assume a human is present, so a purchase that an agent starts often stops halfway.
  • Disputes look like fraud: A charge with no record of consent behind it is hard to defend, so the merchant carries the risk of every unverified agent order.

The problem is not that agents buy. The problem is that agents buy without an identity and without proof of consent.

What does a merchant need to accept an agent payment?

A merchant needs a verifiable agent identity, evidence that a human approved the specific purchase, and a payment that arrives as an ordinary transaction on rails it already uses.

  1. 1

    A verifiable agent identity: The merchant, or the network in front of it, must be able to check cryptographically which agent sent a request. A name in a user-agent header is not enough, because anyone can copy it.

  2. 2

    An approved purchase: The merchant needs evidence that a specific human approved this basket, at this total, from this merchant. A general permission to spend does not answer that question.

  3. 3

    A payment on existing rails: The payment should reach the merchant as an ordinary card transaction, with normal refunds and disputes, so the merchant does not have to build a second payment stack for agents.

How can a merchant verify which agent is calling?

A merchant verifies an agent by checking a cryptographic signature on each request against a public key that the agent operator publishes. Web Bot Auth, an IETF draft that Cloudflare proposed, uses HTTP Message Signatures for this purpose, so a site or its CDN can confirm that a request came from the operator it claims. Visa's Trusted Agent Protocol, developed with Cloudflare, builds on the same signature approach and adds commerce context to the signed request.

A verified identity does not make an agent trustworthy by itself. It lets the merchant apply a policy to a known operator, such as allowing it, rate limiting it, or blocking it, instead of guessing from traffic patterns.

How does a merchant know the user approved the purchase?

A merchant knows the user approved a purchase when the payment carries a signed record of the specific items, total and merchant that the user confirmed. Several standards define such a record. the Agent Payments Protocol (AP2), which Google created and donated to the FIDO Alliance describes signed mandates that capture what the user asked for and the cart the user approved. Visa Intelligent Commerce and Mastercard Agent Pay tie agent transactions to tokenized credentials and to authentication by the cardholder, such as a passkey.

That record matters when a dispute opens. A merchant that can show what the user approved, when they approved it, and what was charged is in a stronger position than a merchant that can show only an automated order.

How does a merchant get paid without a new checkout?

A merchant gets paid without a new checkout when the agent payment arrives as a normal card or wallet transaction through the merchant's existing processor. The details depend on how the merchant sells.

How you sellWhat you changeHow you get paid
A website or app with a checkoutLittle or nothing. Allow verified agents through bot defenses, and accept the tokenized credentials your processor already supports.Through your existing checkout and processor, on the user's card or a network token for it.
A protocol-based checkoutExpose a checkout endpoint that follows a protocol such as the Agentic Commerce Protocol (OpenAI and Stripe) or the Universal Commerce Protocol (Google and Shopify).The agent platform passes a payment credential that your processor charges.
An MCP server with paid toolsSet a price or a quote for each paid tool, and require approval of the amount before the tool runs.The user approves the named amount, a payment layer charges the user's card, and your tool runs.

For an MCP server, a common flow is short. An agent calls a paid tool, and the tool answers with a price instead of running. The agent asks the user to approve, the user approves the named amount, and only then does the tool run. If the work fails, the user should not be charged for it. The merchant keeps its own tool code and decides its own prices.

What stays the same for the merchant?

Most of a merchant's payment operations stay the same when agents buy, because an agent purchase is still a card transaction under card network rules.

  • Card numbers can stay out of your systems: Network tokens and tokenization by a PCI DSS compliant provider keep the raw card number away from the agent and from your agent-facing systems.
  • The charge is an ordinary card transaction: Refunds and chargebacks follow the paths you already know.
  • You control what agents can do: You decide which actions are open to guests, which need a sign-in, and which cost money.
  • The user stays in control: A well-designed agent payment flow lets the user revoke an agent at any time and makes no purchase without approval.

What should a merchant do first?

  1. 1

    Decide where agents meet you: Agents may reach you through your website, through a protocol checkout endpoint, or through an MCP server that exposes your actions as tools.

  2. 2

    Review your bot defenses: Find out whether your CDN or bot management provider can verify signed agent requests, and set a policy for verified agents.

  3. 3

    Choose what agents may do: Decide which actions are open to anyone, which need the user to sign in, and which cost money.

  4. 4

    Price the paid actions: Set a fixed price, or a quote that your code calculates from the request.

  5. 5

    Test before you go live: Run the full approval and payment flow in a sandbox that charges no real money.

This page is a general overview. It is not legal advice. Dispute outcomes depend on card network rules, issuer policy and the agreements between the parties.

Agent traffic is blocked today because it is anonymous and unapproved. Give the merchant a verified agent, a purchase the user approved, and a payment on rails it already trusts, and the agent becomes a customer the merchant can accept.

Frequently asked questions

How can merchants accept payments from AI agents?

A merchant can accept an AI agent payment when it can verify the agent with a cryptographic signature, see a record that a human approved the specific purchase, and receive the payment as a normal card transaction through its existing processor.

Why do websites block AI shopping agents?

Websites block AI shopping agents because bot defenses cannot tell a legitimate buying agent from a scraper or a fraud script, so they block unidentified automation. An agent with a verifiable identity and proof of user consent is something a merchant can decide to trust instead.

Do merchants need to rebuild checkout to accept agent payments?

Usually not. A merchant with a website can often keep its existing checkout and processor and allow verified agents through. A merchant can also add a protocol checkout endpoint, such as one that follows the Agentic Commerce Protocol or the Universal Commerce Protocol, or price the tools on an MCP server.

What standards exist for agent payments?

The main agent payment standards include Visa Intelligent Commerce, Mastercard Agent Pay, the Agentic Commerce Protocol from OpenAI and Stripe, the Universal Commerce Protocol from Google and Shopify, the Agent Payments Protocol (AP2), which Google created and donated to the FIDO Alliance, Visa's Trusted Agent Protocol, and Web Bot Auth for agent identity.

What happens if a customer disputes an agent purchase?

A customer disputes an agent purchase through their bank, as with any card purchase. A merchant that holds a record of what the user approved, when, and what was charged can use that record to answer the dispute.

Do merchants see card numbers from agent purchases?

Merchants do not need to see card numbers from agent purchases. Network tokens and tokenization by a PCI DSS compliant provider let the charge run without exposing the raw card number to the agent or to the merchant's agent-facing systems.

Related articles