.jpg)
Mastercard Agent Connect is the card network’s latest attempt to make AI agents part of everyday commerce, giving merchants a single connection through which autonomous shopping tools can discover products, build carts and complete consumer-authorized payments. Announced on Sept. 9, the system is designed to solve a problem that is becoming more important as AI moves from recommending what people should buy to actually carrying out parts of the purchase on their behalf.
The idea is straightforward, even if the infrastructure behind it is not. A merchant should not have to build and maintain a separate integration for every AI assistant that wants to interact with its store. Mastercard wants Agent Connect to become the common layer between merchants, AI agents, commerce platforms and payment providers.
That puts Mastercard in an interesting position. It is not trying to build the AI model that decides which shoes, flight or laptop someone should buy. It wants to control the trusted infrastructure that allows the decision to become a transaction.
Mastercard Agent Connect Creates One Route Into AI Commerce
Today’s online shopping infrastructure was built around humans browsing websites and apps. Agentic commerce changes that interface.
Instead of opening several product pages, a consumer could tell an AI agent to find a particular product within a budget, compare the available options and prepare the best one for checkout. In more advanced scenarios, the agent could carry out much of that process automatically within limits established by the customer.
Mastercard Agent Connect is designed to give participating AI agents access to merchant-authorized product information, including descriptions and availability. An agent can then recommend products, assemble a cart and confirm details such as final pricing, taxes, shipping and fulfillment before moving toward payment.
That sounds similar to a conventional e-commerce API, but the network model is the important part.
Rather than merchants establishing individual connections with potentially dozens of AI platforms, Mastercard is offering a single integration point connecting participating merchants, agents, commerce services and payment providers.
For merchants, that reduces integration work. For AI companies, it potentially creates a larger pool of stores that agents can interact with through a standardized commerce layer.
Merchants Keep Control of Their Products
One of the biggest challenges with AI shopping is not actually payment. It is making sure the AI has accurate information before it attempts to pay.
An agent recommending an item based on yesterday's price or trying to order something that sold out an hour earlier creates a poor experience for everyone involved.
Agent Connect therefore gives merchants control over the product information exposed to participating AI experiences. Businesses can decide which AI environments they want to connect with while retaining control over pricing, fulfillment, branding, business rules and customer relationships.
That control becomes increasingly important if shoppers begin spending less time on merchant websites.
In traditional e-commerce, the retailer controls the product page, recommendation engine and checkout flow. In agentic commerce, some of those interactions may happen inside an AI interface instead. The AI could become the place where the customer discovers the product, compares it with alternatives and makes a decision.
The merchant then risks becoming invisible infrastructure behind someone else's interface.
Mastercard's approach attempts to preserve the merchant's role while still making its catalog available to AI agents.
Agent Pay Handles the Authorization Problem
Product discovery is only half of the challenge. Allowing software to spend a person's money creates a much more sensitive problem: proving that the customer actually authorized the purchase.
Mastercard is connecting Agent Connect with Mastercard Agent Pay, its infrastructure for agent-initiated payments.
Agent Pay uses Mastercard's Verifiable Intent framework to help establish that a transaction carried out through an AI agent reflects the consumer's authorization. Mastercard also says Agent Connect can support secure payment credentials from any network, while additional network-agnostic tokenization capabilities could be introduced later.
This distinction matters.
There is a significant difference between telling an AI, “Find me the cheapest direct flight to Singapore,” and giving that same software unrestricted access to a credit card.
Agentic commerce needs permission boundaries: what the agent can purchase, how much it can spend, under what conditions it can act and when the consumer must approve the transaction.
Payment networks already specialize in authentication, fraud controls and determining who bears responsibility when something goes wrong. AI shopping introduces a new participant into that chain — autonomous software — but does not remove those existing requirements.
Anthropic Gives Merchants a Route to Build Their Own Agents
Mastercard is also expanding its Agent Suite for Merchants through work with Anthropic.
The company plans to make Anthropic's commerce agent blueprint available through the suite, allowing merchants to develop shopping experiences using Claude models alongside Mastercard's commerce intelligence, controls, personalization and payment capabilities.
That is an important part of the strategy because Mastercard is not assuming all AI shopping will happen through large third-party assistants.
Some retailers may want their own branded agents embedded directly into their websites or apps. Those agents could help shoppers research products, compare options and move toward a purchase while the merchant retains more control over the relationship.
The broader Agent Suite is intended to support both sides of that market: helping merchants become accessible to outside agents through Agent Connect while also giving businesses tools to build agent-powered experiences on their own channels.
Why a Card Network Owns Agentic Commerce
The skeptical question is why any of this needs Mastercard, and the answer is that trust at checkout is a regulated, liability-absorbing product that nobody in the AI stack wants to build. Model companies will not underwrite fraud. Merchants will not accept orders from an agent without a guarantee the payment clears. Whoever standardizes agent authorization becomes the toll booth for a channel that, by most projections, handles a meaningful share of product discovery within a few years, the same interstitial position that stablecoin business accounts and the 21-bank consortium building its own token are fighting for. The banks' consortium play and Mastercard's agent rails are converging on the same thesis from opposite ends: programmable dollars need an authorization standard, and the network that writes it wins the flow.
Two honest caveats. Catalog synchronization is the quiet failure mode: agents submitting orders against stale prices or out-of-stock inventory produce checkout errors and churn at machine speed. And permission boundaries are legally unsettled, an agent authorized to buy within a budget has not been authorized to upsell itself, and where that line sits is currently being written by whoever ships the tokens. The AI venture wave keeps funding competitors, but distribution is the moat here, and Mastercard's acceptance network is the largest in the world.
Mastercard Is Building Before AI Shopping Becomes Mainstream
Mastercard is effectively betting that agentic shopping will become large enough to require its own transaction infrastructure.
Its own recent research points in that direction. A Mastercard report published this week projects that more than one in ten online shoppers could routinely use AI agents to make purchases on their behalf by 2030. That is a forecast rather than a guarantee, but it helps explain why payment companies are building the rails before consumer behavior fully changes.
Agent Connect is initially being rolled out from the United States, with companies and platforms including Samsung, Trip.com, Global Payments, Nexi and others exploring commerce experiences using the network. Mastercard says the platform will continue gaining capabilities as the market develops.
The strategic battle is therefore bigger than AI-powered product recommendations.
If agents increasingly become the interface through which consumers discover and purchase goods, someone still has to connect the AI's decision to a merchant's inventory, verify the shopper's permission and move the money securely.
Mastercard wants that layer to look a lot like the payment network it already operates — except this time, one side of the transaction may be software acting for a human.
