Visa Trusted Agent Protocol and Mastercard Agent Pay: What Merchants Need to Accept AI Agent Payments

The Card Networks Have Entered the Agent Conversation

Most merchant conversations about agentic commerce start with the AI platforms: ChatGPT checkout, Google's protocols, Perplexity, Claude. That makes sense, because those are where shoppers ask for recommendations. But when an agent actually pays, the money still moves over the card networks. And Visa and Mastercard have each built their own programs to decide which agents are trustworthy and how agent payments should look when they reach your checkout.

This matters for a practical reason. An AI agent that is recommended, picks your product, and then gets blocked by your bot protection or declined by your fraud rules is a lost sale you will never see in your analytics. The card-network layer is where a lot of those silent failures will happen. If you have already worked through the protocol layer in our ACP vs AP2 guide, this is the next layer down.

Visa: Trusted Agent Protocol and Intelligent Commerce

Visa's agent program has two halves: one that identifies the agent, and one that handles the credential the agent pays with.

Trusted Agent Protocol (TAP). Visa announced TAP on October 14, 2025, developed with Cloudflare. Its stated purpose is to help merchants recognize AI agents that carry real purchase intent and separate them from malicious automation. According to Visa's announcement, a trusted agent can pass three kinds of information, each carried with agent-specific cryptographic signatures:

  • Agent intent — a signal that the agent is trusted and is trying to view or buy a specific product.
  • Consumer recognition — whether the shopper behind the agent already has an account or history with your store.
  • Payment information — optional payment data, so you can keep using your preferred checkout method.

TAP is built on the HTTP Message Signatures standard and aligned with Web Bot Auth, the same signing mechanism we covered in our post on signed agents and Web Bot Auth. Visa lists the problems it is meant to solve: bot detection blocking legitimate agent purchases, supporting both guest and logged-in agent checkout, and keeping merchants' visibility into consumer and payment data. Visa named Adyen, Checkout.com, CyberSource, Elavon, Fiserv, Microsoft, Nuvei, Shopify, Stripe, and Worldpay among its early partners. The initial specification applies to the Visa network.

Visa Intelligent Commerce. This is the credential side: agents pay with tokenized Visa credentials rather than raw card numbers, tied to authentication the cardholder has already completed. Through 2026 Visa has expanded the program. In April it announced Intelligent Commerce Connect, a merchant-facing product in pilot that Visa says supports agent-initiated payments arriving through protocols including ACP and UCP. In July it reported live agentic transactions in Europe with participating issuers and merchants, using Visa Payment Passkeys for authentication.

Mastercard: Agent Pay and Agentic Tokens

Mastercard's program, Agent Pay, follows a similar logic with different terminology.

Agent registration first. Mastercard's Agent Pay Acceptance Framework, published October 14, 2025, starts by registering and verifying AI agents before they are allowed to transact on its network. Unregistered agents are not supposed to be presenting Mastercard agentic credentials at all.

Web Bot Auth at the CDN. Mastercard partnered with Cloudflare to support Web Bot Auth in Agent Pay. Mastercard says this lets merchants verify agent authenticity at the CDN layer without deploying new code, which is the most merchant-friendly part of either program if your CDN supports it.

Agentic tokens in normal card fields. Trusted agents submit what Mastercard calls a Dynamic Token Verification Code, an agentic token formatted to fit standard card payment fields. In plain terms, the agent can complete your existing checkout form, and the transaction arrives as a tokenized Mastercard payment rather than requiring a new API.

Purchase intent data. The framework includes intent data such as cart contents, transaction limits, and validity windows. Mastercard notes this gives merchants an audit trail that may help avoid or resolve cardholder disputes, which is the part most merchants should care about most.

Why Your Store Might Be Turning These Agents Away

Neither program helps you if your stack treats every non-human visitor as a threat. The common failure points:

  1. Bot protection that blocks before verification. If your WAF, bot manager, or app-level CAPTCHA fires on headless traffic before it checks signatures, a verified agent never reaches the product page. Both Visa and Mastercard built on Web Bot Auth precisely so this check can happen at the edge, but only if your provider has it enabled.
  2. Fraud rules tuned for human sessions. Agent checkouts look unusual by human standards: no mouse movement, very fast form completion, sometimes a new device fingerprint every time. Velocity and behavioral rules written for humans can flag them even when the payment credential is a network-issued agentic token.
  3. Checkout friction agents cannot handle. Forced account creation, interstitial popups, and custom JavaScript-only payment widgets all raise the odds that an agent abandons, regardless of how trustworthy it is.
  4. No way to see what happened. If your analytics and order system do not flag agent-initiated orders, you cannot tell whether agents are succeeding, failing, or never arriving.

What to Do This Quarter

You do not integrate with Visa or Mastercard directly. For nearly every merchant, this work runs through your payment service provider, your platform, and your CDN. A practical sequence:

  1. Ask your PSP three questions. Do you support Visa Intelligent Commerce and Mastercard Agent Pay agentic tokens? How are agent-initiated transactions flagged in my transaction data? Do your risk rules treat network-verified agent payments differently from anonymous card-not-present traffic? Stripe, Adyen, Checkout.com, Worldpay, and others are named partners in these programs, so your provider should have a concrete answer.
  2. Check your CDN and bot manager for Web Bot Auth. Confirm signed-agent verification is on, and that verified agents are allowed through to product, cart, and checkout paths. Re-check any custom rules you added during past bot attacks.
  3. Review fraud rules for agent traffic. Work with your fraud team or provider to make sure behavioral signals alone do not decline transactions that carry network agentic tokens. Do not loosen rules broadly; scope exceptions to verified credentials.
  4. Simplify checkout for machines. Guest checkout, standard form fields, and clear error messages help agents and humans alike. Our technical foundation work covers the rendering and markup side that agents depend on before they ever reach checkout.
  5. Keep dispute evidence. If your PSP passes through agent intent data, make sure it is stored with the order. Mastercard positions that data as dispute evidence, and you will want it when a shopper says they did not authorize what their agent bought.
  6. Start measuring. Tag agent-originated sessions and orders where your platform exposes the signal, and track recommendation-to-purchase outcomes over time with AI visibility tracking.

The Bigger Picture

The card networks are converging on the same shape: verify the agent cryptographically, give it a scoped token instead of a card number, and pass along proof of what the shopper authorized. That shape fits on top of ACP, AP2, and UCP rather than competing with them. For merchants, the takeaway is straightforward. Being recommended by AI gets you considered. Accepting a verified agent's payment without friction is what turns that consideration into revenue, and the card-network layer is where you make sure the door is actually open.

Want to see how AI engines perceive your brand?

Get an Audit for $87