Can AI shopping agents buy from your online shop?

Yes, if your shop runs on Shopify or Commerce.com, and probably not yet if it is a custom built site. Since 28 September 2026, AI agents working inside a shopper's browser can search Shopify stores, fill baskets and complete checkout once the buyer approves. A custom site needs a small amount of extra work, using a proposed browser standard called WebMCP, before agents can do the same reliably.

What changed in September and October?

Two big platforms opened their shops to AI agents within a month of each other. On 28 September 2026, Shopify said browser based AI agents can now complete purchases on its merchants' sites. Before that, agents could search products and add items to a basket, but not pay.

Shopify did this by adding three checkout tools, get_checkout, update_checkout and complete_checkout, including support for Shop Pay. An agent can read the checkout, change the delivery address or delivery option, and place the order once the buyer authorises it. According to TechCrunch, it is rolling out to all eligible Shopify merchants.

On 21 October 2026, Cloudflare announced that Commerce.com had switched WebMCP on across the almost 100,000 store domains it runs through Cloudflare. Commerce.com is the company formerly known as BigCommerce . On those stores, agents can search products and add them to a basket, and the shopper still completes checkout themselves.

What is an AI shopping agent, in plain terms?

An AI shopping agent is an assistant that does the clicking for a shopper. Someone asks it to find a pair of walking boots under £120 in size 9, and it searches shops, compares options and puts one in the basket, sometimes going as far as paying.

Until recently, agents did this by looking at a web page the way a person does: reading screenshots, guessing which button is "Add to basket" and hoping the layout does not change. That is slow and error prone. A product page with a cookie banner, a pop up and a size picker can easily defeat it.

The change this autumn is that shops can now hand agents a proper menu of actions instead. "Search products", "add to basket" and "update checkout" become named tools the agent can call directly, with clear inputs, rather than buttons it has to find.

What is WebMCP?

WebMCP is a proposed browser standard from Google that lets a web page publish a list of tools an AI agent in the browser can use. The tools run inside the shopper's own browser session on your domain, so nothing like a password is handed to a third party.

Google's documentation for developers describes two ways to add tools. The declarative approach works by adding a few attributes to an ordinary HTML form, such as toolname and tooldescription, so an existing search box or enquiry form becomes something an agent can fill in. The imperative approach registers tools in JavaScript for anything more involved, such as a basket or a booking flow.

It is worth being honest about how early this is. Google's own page says WebMCP "is under active discussion and subject to change", and Cloudflare describes it as an experimental proposed standard in early preview, available today through Chrome flags, Chrome origin trials and a handful of agent focused browsers. It is real, but it is not yet something most shoppers' browsers use by default.

How many UK shoppers actually use AI to shop?

A minority today, but a large and interested one. In a survey of 1,000 UK online shoppers carried out in March and April 2026 for Commerce and PayPal , 21 percent said they already use AI tools for online shopping and 64 percent said they were interested in trying agentic shopping tools.

The same survey shows where the hesitation sits. 43 percent were concerned about an AI buying something without their approval, and 83 percent expect AI shopping tools to offer payment security at least as good as what they use now. That is a vendor commissioned survey, so treat the exact numbers as a direction rather than a measurement.

The practical reading for a small shop is simple. Agent shoppers will not be most of your Black Friday traffic on 27 November. They are, however, likely to be the fastest growing slice of it over the next year, and the shops that are easy for an agent to use will be the ones an agent recommends.

What should a small shop on Shopify or Commerce.com do?

A small shop on either platform mostly needs to check its product data, because the platform is doing the plumbing. The agent can only act on what your store tells it, so the quality of your catalogue now decides whether you are chosen.

Before Black Friday, it is worth an afternoon on these:

• Product titles that say what the thing is. "Women's waterproof walking boot, brown leather" beats a clever product name an agent cannot interpret.
• Accurate stock, sizes and variants. An agent that adds an out of stock size to a basket gives the shopper a bad experience, and that shopper blames you.
• Delivery options and costs stated clearly. Shopify's checkout tools let an agent change the delivery option, so the options need to be sensible and correctly priced.
• Returns and terms in plain text. Agents and the assistants behind them read policy pages to answer questions like "can I return this".

If you sell on Commerce.com and would rather not take part, Cloudflare says each merchant can opt out from the Commerce.com dashboard. TechCrunch did not describe a Shopify opt out, so check your admin settings if that matters to you.

What does a custom built shop need to join in?

A custom built shop needs its key actions exposed as WebMCP tools, because there is no platform switching it on for you. For most sites that is a modest piece of development work, not a rebuild.

In practice, that usually means three steps:

1. Search first. Turn your product search into a tool, which for a well built site can be as small as adding the declarative attributes to the existing search form.
2. Then the basket. Register an "add to basket" tool that takes a product and a quantity, using the same code your buttons already call.
3. Checkout last, if at all. Letting an agent pay is the most sensitive step. Commerce.com stops at the basket and leaves payment to the shopper, which is a sensible place for a small business to stop too.

There is a second option for sites served through Cloudflare. In the same post, Cloudflare explains that its edge feature injects a small script so WebMCP can work without the site owner writing code, provided the shop already has an MCP server, a back end service that describes its tools. That suits some setups and not others, which is exactly the kind of decision worth a short conversation before anyone writes code.

If your shop is a custom web app built for your business , the same groundwork that makes it fast and accessible makes it easy for agents too: real HTML forms, clear labels, and product data stored in a structured way rather than baked into images.

Should you let agents in at all?

Not every retailer is saying yes. TechCrunch reports that Amazon is blocking Meta's Muse agent from its retail site and that Adidas is reportedly blocking AI agents from making purchases. Big brands with their own apps and loyalty schemes have reasons to keep shoppers on their own turf.

A small independent shop is in a different position. Most small shops win on being found, not on keeping people inside an app, and an agent that can use your site is one more route for a shopper to find you. The risk is mainly operational: wrong stock data or confusing delivery options become visible faster when software is doing the buying.

A reasonable middle path is to let agents search and fill baskets now, and keep payment in the shopper's hands until the standard settles and you can see how real customers use it.

If you are not sure whether your shop is ready, or what it would take on a custom site, book a free 15 minute call and we will tell you straight, even if the answer is that your platform already covers it.

Frequently asked questions

Do I need to do anything if my shop is on Shopify?

Shopify says agent checkout is rolling out to all eligible merchants, so the technical side is handled for you. What you control is the quality of your product titles, stock levels, variants and delivery options. Agents rely on that data to decide what to recommend and buy.

Is WebMCP safe for my customers' payment details?

WebMCP tools run in the shopper's own browser session on your website, so the agent does not receive the shopper's password or card details directly. On Shopify, the order is only placed once the buyer authorises it. On Commerce.com stores, the shopper completes checkout themselves.

Which browsers support WebMCP today?

WebMCP is still experimental. Cloudflare says it is available through Chrome flags, Chrome origin trials, some agent focused browsers and developer tools. If a browser does not support it, the site simply works as normal for the shopper.

Is this the same as ChatGPT or Google recommending my products?

Not quite. Assistants recommending products relies on your site and listings being readable, while WebMCP lets an agent take actions on your site, such as searching and filling a basket. Good product data helps with both.