WebMCP on Shopify: what AI agents can do on your store

Since August 2026, every Shopify store running a Liquid theme has been exposing 10 structured tools to any AI agent running in the shopper's browser: no app install, no admin toggle, no opt-in required. The tools cover catalog search, cart management, and checkout navigation. They do not cover order status, returns, or anything that completes a purchase.

Understanding what those 10 tools do, and what they cannot do, tells you exactly where your store is ready and where it is not.

What changed in August 2026

WebMCP (Web Model Context Protocol) is a draft web standard, developed in a W3C community group with editors from Google and Microsoft, that lets a page register structured tools with the browser. An AI agent running in the same tab can call those tools instead of scraping HTML and guessing at which button to click.

Shopify announced WebMCP support on August 5, 2026 and began rolling it out to Liquid Online Store 2.0 storefronts. The tools are loaded by the script at cdn.shopify.com/storefront/webmcp/webmcp-0.1.0.js. Nothing in Shopify's current documentation describes a way to modify the default tools, add your own through the Liquid layer, or turn the feature off.

This is not the same as Shopify Agentic Storefronts, which pushes your catalog into ChatGPT and similar surfaces so shoppers can buy without visiting your site. WebMCP serves an agent the shopper already brought with them. The shopper is on your storefront. Their agent is helping them while they are there.

The 10 default tools

Tool	What an agent can do with it
`search_catalog`	Search products, collections, articles, and pages by keyword
`browse_store`	List collections, or browse the products inside one
`get_product`	Pull full product detail: variants, prices, which combinations are in stock
`show_variant`	Navigate to a product page with a specific variant selected; resolves partial matches like "the blue one"
`get_cart`	Read current cart line items and totals
`update_cart`	Add items, change quantities, remove lines; returns clarifying options if the request is ambiguous
`cancel_cart`	Empty the cart entirely
`proceed_to_checkout`	Verify the cart is not empty, then navigate to checkout
`manage_orders`	Send the shopper to their order history; prompts a login if they are not signed in
`search_shop_policies_and_faqs`	Answer questions about returns, shipping, hours, and services from your store's own pages

Every one of these tools reads from or writes to data your store already holds. None of them completes a purchase. proceed_to_checkout navigates to the checkout page; a human still enters payment details and places the order. Shopify drew a firm line before money moves.

What agents still cannot do

The gaps matter more than the capabilities.

Order status (WISMO). manage_orders routes the shopper to their order history page. It does not answer "where is my order?" directly. An agent has no default tool to look up a specific order number, surface the current shipping status, or give an estimated delivery date. This is the most common post-purchase question brands receive, and it is not covered.

Returns and exchanges. There is no default tool to start a return, check eligibility, or generate a return label. If a shopper asks their agent to start a return, the agent has no structured path.

Thin product data. Agents can read the words in your titles, tags and descriptions, but they can only filter on structured fields such as product options. A vendor analysis of 400,000-plus live Shopify products (atomz, September 2026) found that 51% of what shoppers ask about sits somewhere in readable text, and only 6% of that is filterable. If colour is a word in the title instead of a real option, an agent can see it but can't reliably filter by it.

Cart display in custom themes. Shopify's cart tools call the same standard storefront actions (Shopify.actions) that apps use. Themes whose cart code bypasses those standard actions (cart drawers built on custom state management, for example) can show a stale cart to the shopper after an agent edits it. The session cart changed; the visible UI did not.

Stable Hydrogen. Production Hydrogen stores did not receive the browser tools in the August rollout. The Hydrogen developer preview includes them and can turn them off with webMcp={false}. Stable production Hydrogen has a separate server-side endpoint (/api/mcp) for remote agents, which is different.

Late tool registration and overlays. Ora's public WebMCP audit of Reebok's storefront (September 2026) found that the checkout tool registered at 3.6 seconds, while agents typically read the page at roughly 3 seconds. Tools registered after the agent reads the page may not be seen. The same audit found popups and overlays covering page content, and tool descriptions written at excessive length. These are patterns any store can land in.

How to see it yourself

Two options, both free:

Chrome flag and inspector. Open Chrome 146 or later. Navigate to chrome://flags and enable #enable-webmcp-testing. Install the "Model Context Tool Inspector" browser extension. Visit your storefront. The extension lists every registered tool and lets you call them manually.
ChatGPT desktop. Open the ChatGPT desktop app (not the web app). Switch to GPT-5.6 Sol or Terra: these are the only models with WebMCP enabled as of August 25, 2026. Luna does not support it. Enterprise and Edu accounts do not have access. Visit your storefront in the built-in browser. An arrow appears in the address bar when tools are available.

What to do before Black Friday

Black Friday 2026 falls on November 27. Chrome 157, the release targeted for WebMCP on by default, is expected around November 3, 2026, though that timing has not been committed. Browser coverage will still be limited on the day, but the preparation work below improves your store for shoppers regardless.

Structured product data

Go through your best sellers and fix three things. Make colour and size real product options, not words in a title or a tag. Replace vague option values like "Style A" and "Style B", which give show_variant nothing to resolve against. And keep inventory counts accurate, because stale counts produce confident wrong answers.

Policy and FAQ pages

search_shop_policies_and_faqs reads from your store's actual pages. If your return policy, shipping times, or holiday cutoff dates live in a helpdesk macro, a PDF, or an email template, and not on a crawlable store page, the tool cannot find them. Put the operative fact first: return window in the first sentence, shipping cutoff in the first sentence.

Cart behavior

Test whether an agent adding or removing a cart item correctly updates what the shopper sees in the tab. If your theme manages cart state independently of Shopify's standard actions, you will need to reconcile that before agents begin using these tools at scale.

Custom tools for the gaps

Shopify provides no documented way to add custom tools to the Liquid layer. For order status, WISMO answers, and returns, the three gaps that drive the most support tickets, custom tools registered with document.modelContext.registerTool() are the path. They need JavaScript running on your storefront pages and a developer to build them. Because Shopify doesn't document how extra tools sit alongside its defaults, test them on your own theme before relying on them.

What agent traffic to expect

One store owner who added WebMCP tooling to a live eSIM store reported zero agent-placed orders as of September 2, 2026. Current agent traffic through these tools is small. Anyone describing WebMCP as an immediate revenue channel is getting ahead of the data.

The reason to prepare now is different. Adobe reported (Q1 2026) that AI-referred traffic to US retail sites grew 393% year over year and converts 42% better than other traffic. The shoppers arriving via AI surfaces are already engaging differently. When those same shoppers use an in-browser agent on your storefront, clean structured data, working cart behavior, and clear policy pages determine whether the agent helps or confuses. Those foundations also improve your store for every other shopper.