
Shopify is now letting any AI agent into its product catalog of more than a billion listings, and into its checkout, with no partnership deal required. On the same day, American Express unveiled a framework meant to capture what a customer actually told their agent to do. Taken together, the two announcements from October 13 show how agent-driven shopping is supposed to become practical: open for search, strict at payment.
Key takeaways
- Shopify is opening its catalog of more than a billion listings to any AI agent; according to its developer site, search is free and needs neither an account nor an API key.
- Payment runs through Shop Pay, and the shopper confirms every order; at stores outside Shopify, the agent only receives card details for the one approved purchase, including a cap on the amount.
- American Express is introducing “Intent Intelligence,” a process that captures and confirms the purchase instruction before payment so that mismatches get caught.
- For shoppers in Germany, both are still a preview: Shopify publishes no country list, and Amex gives no launch date.
What Shopify is opening up
Until now, AI agents reached Shopify merchants only through individual partnerships. In September, for instance, Shopify agreed that Meta’s agent Muse could shop in Shopify stores and pay via Shop Pay. Now all a shopper has to do is connect their agent to their Shop account. Shop is Shopify’s own platform that consolidates orders, tracks packages, and recommends products. Payment happens with Shop Pay right inside the conversation with the agent, and the order then shows up in the Shop app. Shopify pitches the whole thing as a “personal shopper” that users can give rules, such as a budget or a preference for fast-shipping items.
Under the hood is an interface called Shop MCP, documented at shop.dev. MCP, the Model Context Protocol, is the widely used standard that lets AI assistants call external tools. Through it, agents can search the catalog, pull product details with variants and stock levels, create carts, and look up a shopper’s orders. Filters cover price, size, color, rating, store, and shipping destination, and image-based search is also supported. Developers additionally get a command-line tool. According to the documentation, anyone using an assistant such as Claude or ChatGPT that can connect MCP tools is signed in automatically through “Sign in with Shop.”
The shopper keeps control of checkout
What stands out is how Shopify secures payment. At Shopify merchants, the purchase runs through a Shop checkout. The most detailed documentation covers merchants outside Shopify whose listings are also in the catalog: the agent first sees the merchant’s real terms: items, shipping, tax, and total. It then submits an approval request. The shopper opens a link, reviews the purchase in Shop, and confirms the merchant, the card, and a cap on the amount. Only then does the agent receive payment credentials, and only for that exact purchase on that exact checkout page. It never sees the shopper’s actual credit card number. The documentation also explicitly requires that card numbers and security codes never appear in chats, logs, or error messages.
So this is no blank check for unlimited shopping. According to The Verge, every single order has to be approved; the documentation does not describe standing budgets that span multiple purchases. The Shop sign-in also draws a clear line: the connected app receives name, email, addresses, and orders, while sizes and personal preferences stay with Shop.
Amazon shows how contentious the topic is. The company has blocked Meta’s Muse from its website since September 20, arguing that the agent does not identify itself while browsing. Amazon had already sued Perplexity over its shopping agent. Shopify is taking the opposite approach: agents should sign in openly through the interface instead of imitating a person on a web page. That is also a response to a security problem that OpenAI’s agent test on prompt injection made vivid: an agent that reads and operates third-party websites can be manipulated by hidden instructions. A structured interface with approval by the shopper shrinks that attack surface considerably.
Amex wants to lock in the instruction
American Express works one layer down, at the credit card itself. The “Intent Intelligence” framework presented on October 13 is designed to capture purchase intent before the agent pays at all. An interface called the Create Intent API records the instruction along with constraints such as merchant, maximum budget, and product preferences. The cardmember authenticates and confirms that the agent understood them correctly. That produces a unique identifier and a proof token, which gives the agent a digital payment credential instead of the real card number.
Amex offers a concrete example: if a customer orders white Nike sneakers in size 9 and the agent buys Adidas shoes, Amex could stop the transaction. A log is also meant to record which agent authorized what and when, which helps with fraud cases and disputes. “These new transactions will have a consumer, a merchant, and an agent,” Hannah Bozian, who leads agentic partnerships at Amex, told TechCrunch. What is new, she said, is that the approach pulls intent earlier into the transaction flow than other implementations do.
The framework is part of the “Agentic Commerce Experiences” developer kit that Amex introduced in April. At the time, the company also announced purchase protection for errors by registered agents, but left the details open. Amex has not given a launch date for Intent Intelligence and wants to turn it into an industry standard. So far, the company says, it has not seen an increase in disputes from agent purchases.
What this means for shoppers and merchants
For consumers in Germany, this is a look ahead for now. Shopify publishes no country list; the preview of its search filters shows the US, Canada, the UK, and Australia as shipping destinations. Amex names neither a date nor markets. Anyone who wants to experiment with agents can already try the catalog search at shop.dev, with no account and no cost. Developers get a fast way to build product search into their own assistants instead of scraping websites.
For merchants, the math is less clear. Shopify has not yet published figures on whether agent channels bring in additional sales. Anyone selling on Shopify should focus above all on clean product data: an agent chooses based on attributes such as size, color, delivery time, and return policy, not on nicely designed storefronts.
Bottom line: trust becomes infrastructure
The two announcements tackle the same question from opposite ends. Shopify makes finding and ordering as easy as possible for agents but keeps the final sign-off with the human. Amex wants to make what the human intended provable at the card level. That shifts the debate: what matters less is whether an agent can shop, and what matters more is who can prove, in a dispute, what was actually requested. Whoever establishes that chain of evidence as a standard first, retail platform or card network, is likely to determine how open agent-driven shopping ends up being. Amazon’s block on Muse shows that question is far from settled.
Sources
- Shopify: shop.dev – Shop MCP für Apps und Agenten
- Shopify: Dokumentation zu Zahlungsfreigaben
- The Verge: Shopify opens its catalog up to more AI agents
- TechCrunch: Amex introduces a framework to prevent AI agents from making unauthorized purchases
- American Express: Agentic Commerce Experiences
- Adweek: Amazon locks out Meta's Muse in agentic shopping standoff

