Browser Agents Can Now Check Out on Your Shopify Store. Here's What They Read

On August 5, Shopify switched on a set of tools that let AI agents search your catalog, change a shopper's cart and head to checkout. You didn't install anything. According to the developer changelog, the tools went "live today on every Liquid storefront."

On September 28 Shopify finished the job. A second changelog entry added checkout tools, so an agent can now read the checkout, fill it in and submit the order once the buyer confirms. Again, no setup: the tools "don't expose a new API or require merchant configuration."

So if you run a standard Shopify theme, your store already has an entrance built for software. Nobody asked you, and that's mostly fine. The part worth your attention is what an agent finds once it's inside, because it reads your store's data directly instead of looking at your pages the way a person does.

What WebMCP is, minus the jargon

WebMCP is a proposed web standard. Shopify's changelog says Shopify, Google and Microsoft are developing the spec together. Rather than an AI agent "looking" at your product page and guessing which button adds to cart, the page publishes a list of named actions the agent can call directly, and each one returns structured data.

It runs inside the shopper's own browser tab. That's the difference from the hosted integrations (ChatGPT, Gemini, Copilot) that pull from Shopify Catalog. Here a person has your store open, an assistant is helping them, and the assistant works through your storefront in real time.

Two caveats before anyone gets excited. Shopify's WebMCP docs say agent support is "currently limited to Chromium-based browsers," and the changelog calls it "still an emerging standard" running in an origin trial. Volume today is small. But the plumbing went into every store at once, so whenever agent traffic does grow, you won't get a heads-up first.

The tools, and the store data each one reads

This is the useful bit. Each tool reads something you control:

• `search_catalog` searches products, collections, articles and pages, and returns matches with pricing and availability.
• `browse_store` lists your collections or the products inside one.
• `get_product` pulls full product details: variants, pricing, stock status.
• `show_variant` sends the shopper to a product page with a specific variant pre-selected.
• `get_cart`, `update_cart` and `cancel_cart` manage the cart using standard storefront actions, so your cart drawer opens just as it would for a human.
• `search_shop_policies_and_faqs` answers questions about "returns, shipping, and hours of operation."
• `proceed_to_checkout` and `manage_orders` hand off to checkout and order history.

Look at that list as an agent would. It never sees your hero banner or your lifestyle photography. What it gets is titles, variant names, prices, stock numbers, collection structure and policy text. A shopper might say "find me the waterproof one in a men's 10" and the agent has to work that out from your variant options and product data alone.

That's why a variant called "Option 3" or a size chart that only exists as an image costs you more than it used to. A human can squint at a photo. The agent gets whatever text you gave it.

What the checkout tools can and can't do

The checkout docs are more specific than most coverage of this launch. The `update_checkout` tool can set:

• buyer email and phone
• shipping address, pickup location and delivery option
• discount codes
• tax numbers and store credit
• a saved Shop Pay card or billing address

It deliberately ignores line items. Quantity changes happen on the page. Agents also can't type in new card details, so a shopper without saved Shop Pay credentials still has to enter payment info themselves.

Shopify also sets a rule for agent builders: before calling `complete_checkout`, the agent must "show the buyer the current order and total, and get their permission to place it." Agents have to sign their requests with Web Bot Auth, and the docs warn that unregistered agents "may face deprioritization or blocking."

That last point matters more than it looks. Identification is the exact thing Amazon complained about when it blocked Meta's Muse agent in September, saying agents should "identify themselves clearly in browser traffic rather than disguising themselves as human shoppers." Shopify built that requirement into the protocol. So the agents finishing orders on your store are supposed to be signed and known.

Why the discount code line deserves a second look

Agents can apply discount codes. Good: a shopper says "use my welcome code" and it works.

Less good: if you've ever left a generous code floating around (an influencer code that never expired, a 30% "sorry" code from a support ticket two years ago), an assistant that's told to find the best price will eventually go looking for it. Shopify's docs don't say agents hunt for codes on their own. I'd still assume some will try, because coupon-hunting is exactly the sort of chore people hand to a bot.

Before Black Friday, open Discounts and expire anything you wouldn't want applied to every order. You should do this anyway. Agents just make it more urgent.

Where agent shoppers land

Shopify gave a useful number on its Q2 call. Per TechCrunch's report, half of all AI-referred sessions land directly on a product page, which is 2.5 times the rate for traditional search. Shopify also said AI-driven traffic and orders to its stores tripled year over year.

Put that next to WebMCP and the job is clear. Your product page and the data behind it are your storefront for this kind of shopper. Your homepage hardly comes into it.

What we don't know yet

I'd rather be clear about the gaps than pretend the docs answer everything.

• Shopify hasn't said how agent-assisted orders will show in your reports, or whether you'll be able to tell them apart from regular orders.
• There's no merchant setting to switch the tools off. Neither changelog mentions one.
• The docs don't say how checkout UI extensions or custom checkout fields behave when an agent fills the checkout. If you rely on a required custom field (a gift message, a delivery date picker), watch your first agent orders closely.
• The Shop Pay handling is new. As PYMNTS noted, checkout support arrived a week after Tobi Lütke announced Meta's Muse agent could check out with Shop Pay on all Shopify stores. Expect more agents to show up quickly.

What to do this month

None of this needs a developer. It needs an hour with your catalog.

1. Fix variant option names. Every option should say what it is ("Size," "Color," "Width") and every value should be a word a shopper would type ("Navy," "Wide," "10"). No internal SKU codes as option values.
2. Move key specs out of images. If your size chart, materials or dimensions only exist in a JPG, put them in the description or in product metafields.
3. Fill in your store policies. `search_shop_policies_and_faqs` answers from them. Write returns, shipping and contact details in plain text, with real numbers: days, costs, regions.
4. Check stock accuracy. The agent reports availability straight from your inventory. Phantom stock means an agent tells a shopper "yes, in stock" and you cancel the order later.
5. Audit your discount codes and expire the stragglers before November.
6. Make your collections make sense. `browse_store` exposes your collection structure. A collection called "New Stuff 2" doesn't help a machine or a person.

The stores that do well here won't have done anything special for agents. They'll be the ones whose product data was already clean. When the reader can't see your photos, the text has to do all the selling.