What AI agents in your shopper’s browser can do on your Shopify store: I tested Shopify’s WebMCP tools on live stores

Since 5 August, every Shopify store on a regular theme has been quietly offering AI agents a set of tools: search your catalog, read your return policy, add to cart, go to checkout. Nothing to install, and no off switch in Shopify’s docs. I loaded five Shopify brands in Chrome on 6 October and called those tools myself. On three themed stores, 11 tools were live, and Shopify even supplies the Chrome trial token that switches them on, valid until 17 November 2026. The agent answers return questions from your policy text, word for word. The two custom-built (headless) storefronts I checked exposed nothing. And on one big brand, every cart tool failed. If you run a Shopify store, here is what an agent in your customer’s browser can actually do on it, and the three things worth checking.

What Shopify switched on

Shopify’s developer changelog from 5 August is short and direct: “Online stores now expose WebMCP tools that AI agents can call. Agents can search your catalog, manage the shopper’s cart, and go to checkout on the shopper’s behalf, all in the tab they’re looking at.” And on setup: “The tools are live today on every Liquid storefront and on the Hydrogen developer preview. There’s nothing to install or configure.”

WebMCP is not the same thing as the AI channels I mapped earlier this week. ChatGPT, Copilot, Gemini and Meta’s Muse work from Shopify’s catalog, away from your site. WebMCP is for an agent that runs inside the shopper’s own browser tab, on your actual storefront. Google describes it as “a proposed web standard to help you build and expose structured tools for AI agents.” Without it, an agent has to read your page and simulate clicks. With it, your store hands the agent a menu of named actions.

On 28 September Shopify extended this into checkout. Its docs say “Checkout WebMCP lets AI agents in the buyer’s browser read and update the checkout, and place the order after the buyer confirms it.” The buyer stays in the loop: “The buyer sees the same checkout state, handles page interactions such as Shop Pay login or payment challenges, and confirms the order before your agent calls complete_checkout.” Agents are told: “Before you call complete_checkout, show the buyer the current order and total, and get their permission to place it.”

Who can use it today

Shopify is upfront that this is early: “agent support is currently limited to Chromium-based browsers through an origin trial.” An origin trial is Chrome’s way of letting a website try an unfinished feature. Each site normally has to register for it and serve a token, which Chrome started offering from Chrome 149.

This is where the live check got interesting. Merchants don’t register for anything, because Shopify does it for them. On all three themed stores, the page carried five origin-trial tokens issued to cdn.shopify.com. I decoded them. One reads "feature":"WebMCP" with an expiry of 17 November 2026. With that token present, my Chromium 151 browser had the WebMCP interface switched on for the store, with no flags or settings changed. When I used a headless browser’s default identity instead of a normal Chrome one, the tokens were not added at all.

What happens after 17 November is not something I can tell you. Shopify could renew the token, or Chrome could change the feature, which its docs say is “under active discussion and subject to change.” That date is worth a calendar note, not a panic.

What the agent actually sees

The changelog lists the tools by name. I wanted to see what they return, so I called the read-only ones on Tentree’s store.

Your policies, verbatim. Asking search_shop_policies_and_faqs “What is your return policy?” returned Tentree’s refund policy body, which in full is: “$10 Returns (includes shipping label), Free Exchanges. Please note, products with a discount of 30% or more are final sale.” Along with it came an instruction to the agent: “The body is the authoritative policy text” and “do not invent details that are not present.” So the agent’s answer to a return question is only as good as the text in your policy page. If your shipping policy says “see FAQ”, that is what the agent gets.

Your own store search. search_catalog is not a new AI index. Reading Shopify’s script for these tools, the catalog search runs on predictiveSearch, the same type-ahead search behind a store’s search box. Shopify’s API docs say “Product results always search all product fields.” A search for “organic cotton hoodie” on Tentree returned 10 products, the most I asked for. The agent finds what your search box finds, using the words in your titles, descriptions, types and tags.

The cart, with your theme’s behaviour. On Death Wish Coffee, add_to_cart added a bag of coffee and get_cart read it back with the price. Shopify notes that cart tools “call the same standard storefront actions that apps use, so if your theme opens a cart drawer on updates, agents trigger it too.”

Where it broke

On Allbirds, catalog search and the policy lookup worked fine. Every cart call failed with the same error: Variable $l of type LanguageCode! was provided invalid value. I tried three times over about 15 minutes, with the browser set to US English. The two stores where cart tools worked both report their storefront language as en. Allbirds reports en-US. My guess is that a regional language setting trips the cart tools, but that is an inference from one store, not something I can prove from outside. The point for a merchant is simpler: these tools run on your live store, with your settings, and they can fail without you ever knowing, because no shopper will email you to say their agent couldn’t add to cart.

I could not test checkout the same way. On Tentree’s checkout page, an ordinary visitor saw no WebMCP tools. Shopify’s docs say the checkout tools register “on eligible checkouts” and that agents sign their requests with Web Bot Auth, a way for a bot to prove who it is. Which of those applied, I can’t tell from outside.

Three things to check on your store

1. Is your storefront a theme or headless? If a developer built your storefront as a custom app (Gymshark’s runs on Next.js, Ruggable’s on Vercel), none of this applies. On both, I found no WebMCP script and no tools. Your agent story depends on whatever your developer builds.
2. Read your policy pages as if they were the only thing an agent knows about you. Refund, shipping and contact are policy types the tool can look up. Write them in full sentences with the real numbers, because the agent is told to stick to that text.
3. Test your own cart. Chrome’s WebMCP docs point developers to a Model Context Tool Inspector extension. Open your store in a current Chrome, open the inspector, and run search_catalog, add_to_cart and get_cart. If you see an error like the Allbirds one, send it to Shopify support with your store’s language settings.

I have no figure for how many shoppers use a browser agent today, and I haven’t seen anyone publish one for Shopify stores. What has changed is who did the setup work. Shopify has already done it for most stores. The parts left to you are what the agent reads (your policies and product text) and whether the tools actually work on your store.