If you run a fashion store on Shopify, there is a decent chance an AI agent has already been through your product pages this month. Until recently it could look, compare and fill a cart, then it had to hand the tab back to a person. Since the end of September that last step is gone. With Shopify WebMCP, an agent running in the shopper's browser can now read the checkout, fill in the delivery details and place the order once the shopper says yes.
That sounds like news for developers. It is really news for anyone who owns a product page, a size guide or a checkout customisation, because those are the things an agent now reads instead of a person. Here is what changed, what it means in practice, and the checks I would run before the peak season traffic arrives.
What Shopify actually switched on
On 28 September 2026 Shopify announced that browser-based AI agents can complete checkout on merchant stores. Before that, agents on Shopify could search products and build carts, and that was where it stopped.
The mechanism is WebMCP, a proposed web standard for agents that live inside the buyer's browser. Shopify added three checkout tools on top of it: one to read the checkout, one to update it (address, delivery option, discount code, payment) and one to complete it. The Shopify developer documentation also lists a fourth tool that sends the tab back to the storefront. Agents no longer need to take screenshots of your checkout or guess at your HTML. They call structured tools and get structured answers.
Two details matter more than the headline. First, the shopper stays in charge: the agent must get permission for the current order and total before it completes, and has to ask again if the total changes. Second, the rollout is to all eligible merchants, and neither The Paypers nor Shopify's own pages spell out who is not eligible. If your checkout is heavily customised, assume you need to test it rather than assume it works.
How an agent shops your store today
It helps to picture the journey. Someone asks an assistant for a linen shirt under 120 euros that ships to Spain by Friday. The agent searches, and on Shopify that search increasingly runs on structured product data rather than scraped pages. Shopify's Spring '26 Edition says merchants with eligible products are included in Shopify Catalog by default, and that AI searches powered by Catalog convert at twice the rate of searches using scraped data. Treat that as Shopify's own figure, but the direction is clear.
The agent then lands on your product page, reads the title, the variants, the size information and the delivery promise, adds to cart and moves to checkout. With WebMCP it reads the checkout through the tools, sets the address and shipping method, and asks the shopper to confirm. If Shop Pay asks for a login or a payment challenge, the shopper finishes that part in the same tab.
Everything the agent decides is based on what it can read. That is the uncomfortable part. Adobe measured how machine readable US retail sites are and product pages came out lowest of every page type, scoring 66% on its content visibility checker. Homepages scored 75%. The page that should sell the product is the one agents understand least.
Why this is worth an afternoon now
The traffic is no longer a rounding error. In the same April 2026 report, Adobe found AI-referred traffic to US retail sites grew 393% year on year in the first quarter, and in March 2026 it converted 42% better than other traffic. A year earlier it converted 38% worse. That swing is the real story: people who arrive through an assistant now arrive closer to buying.
Shopify is also building the merchant side around it. Agentic Storefronts in the admin let you choose which AI channels can discover your products and check out directly, and show orders and conversions from ChatGPT, Copilot, AI Mode in Google Search and the Gemini app. Sidekick suggests fixes for products that show up in AI conversations but do not convert. None of this needs a new app. It does need clean product data and a checkout that behaves.
What breaks first on fashion stores
From the stores we look at, the problems are rarely technical in the dramatic sense. They are small decisions that made sense for a human and confuse an agent.
Product titles that carry only the collection name are the first one. "The Marina" tells a loyal customer everything and tells an agent nothing. If colour, material and fit live only in an image or a tab that loads after a click, the agent may never see them. The same goes for size guides that are a picture of a table.
The second is anything that interrupts the path. Pop-ups that cover the page on arrival, age or country gates, and cart drawers that need a specific click are built for eyes. The documentation is explicit that agents must never operate the page's controls themselves when the checkout asks the buyer to act, so every interruption becomes a hand-off back to the person.
The third is checkout customisation. The WebMCP tools cannot change line items, only accept saved Shop Pay cards for payment and only support two declared fields, a tax number and store credit. If your checkout depends on a custom field, a gift message that is mandatory, or a pre-order condition the shopper has to tick, the agent will hit a wall there and send the buyer back to finish it by hand. That is not a disaster. It is just a lost sale that you will never see in a report.
Five checks to run this week
You do not need a project to start. You need an afternoon and someone who will be honest about what they find.
Start with your ten best-selling products and read each one as if you could not see images. Is the colour, material, fit and care in the text? Is the size guide text, not a picture? If the answer is no, that is your first fix, and it helps human shoppers and search too.
Then check delivery and returns. An agent asked for "delivered by Friday" can only answer if the delivery promise is written on the product page or in the shipping rates. Vague phrases like "fast shipping" are invisible.
Third, walk your checkout with every customisation switched on and list each step a person must complete. Each one is a potential hand-off. Decide which are essential and which are habits.
Fourth, open Agentic Storefronts in the admin and look at which AI channels are enabled and what the performance view shows. Many merchants have never opened it.
Fifth, look at your pop-ups and gates on mobile and ask whether each one earns its place. If it does not convert people, it will not convert agents either.
What comes next
WebMCP is one piece. Shopify's hosted MCP server handles agents that work server to server, and both run on the Universal Commerce Protocol that Shopify built with Google, which Shopify merchants are enabled for by default. The direction of travel is that agents will discover, compare and buy through structured data, and the store's job is to make that data complete and the checkout predictable.
This is also why we built AI checks into our Shopify AI development process at MM Shopy Devs. When our in-house system reviews a store, product data and checkout behaviour for agents are becoming part of the same QA pass as speed and accessibility, because they fail in the same quiet way.
Final thoughts
The shift here is not that robots will replace your customers. It is that more of your customers will send something to read your store for them, and that something does not care about your hero banner. It reads titles, attributes, delivery promises and checkout steps. Brands that write those clearly will be easier to buy from, for people and for agents. If you want a second pair of eyes on yours, get in touch and we will tell you what we would fix first.
