Universal Commerce Protocol

When you buy something online, you click your way through a page: search for a product, add it to the cart, pay. These pages are built for human eyes. A computer program tasked with doing the same thing struggles with this. It would have to guess where the buy button is and what counts as a price. The Universal Commerce Protocol is an attempt to make this guessing unnecessary. It establishes a common language in which a shop reports its products, prices, and order steps to a program — and in which the program responds back.

Why shops now need a machine checkout

More and more people are letting AI assistants help them shop. You describe what you’re looking for and get suggestions. So far, the last step has usually broken off: the assistant sends you to a website, and there you buy it yourself. It is precisely this gap that a commerce protocol is meant to close.

For merchants, a lot is at stake here. If assistants become the primary shopping interface, a shop without a machine-readable interface is practically invisible. It simply won’t show up in the assistant’s suggestions. This explains why major commerce and payment companies are collaborating on such standards rather than leaving them to a single AI provider.

A standard also has a practical advantage. Without one, every shop would have to build its own integration for every assistant. With five assistants and a thousand shops, that would mean five thousand custom solutions. With a shared protocol, each side builds its integration once.

From shopping cart to purchase approval

Technically, such a protocol consists of defined messages between two programs. The shop offers an interface, i.e. an address to which other software can send requests. The assistant asks, for example: Which running shoes in size 43 do you have under 100 euros? The shop responds in a structured data format, not as a colorful webpage. Every field has a fixed meaning: item number, price, availability, shipping costs.

This is followed by further defined steps: creating a cart, adding items, submitting a delivery address, confirming the order. You can think of it like a form whose fields are named the same across all shops. The assistant no longer needs to know what a particular shop looks like. It only needs to master the protocol.

The most delicate part is payment. No merchant wants a program spending money without control. That’s why such protocols work with tightly limited payment authorizations. The user, for instance, permits exactly one payment up to a fixed amount at a specific merchant. In addition, there is a marker indicating that an assistant is acting on behalf of a human — this matters if someone later disputes an order.

UCP, MCP, and the fight over the standard

The name Universal Commerce Protocol has been appearing since 2025 in reports about Google and its partners from commerce and payments. In parallel, there are competing proposals, such as the Agentic Commerce Protocol from the circle around OpenAI and Stripe. It is currently open whether one of them will prevail or whether several will coexist. This is exactly what business media are reporting on when they write about the race for the standard in agentic commerce.

A common misconception is confusing this with the Model Context Protocol, or MCP for short. MCP generally governs how an AI model addresses external tools and data sources. A commerce protocol is narrower and specifically describes the process of a purchase. Both can work together, but they don’t mean the same thing.

In everyday life, you will never see the protocol itself. It becomes noticeable in a different way: an assistant shows you prices and delivery times directly in the chat and completes the purchase after your confirmation. To you, this looks like a single button press. Behind it runs a chain of coordinated messages between assistant, shop, and payment provider.