Developer Changelog / Storefronts

WebMCP support
Stores now expose their own "control surface for AI agents"

Online stores can now expose WebMCP tools, letting AI agents search the catalog, work with the cart, and move to checkout directly inside the tab the shopper is looking at. Live as of today across all Liquid storefronts and in the Hydrogen Developer Preview. Nothing to install or configure.

On this page
  1. What actually changes (in 30 seconds)
  2. How it works: traditional agents vs. WebMCP
  3. The tools agents can call
  4. Click simulation vs. WebMCP
  5. Coverage and prerequisites
  6. What merchants need to do (3 steps)
  7. 5 things developers should know
  8. 3 practical use cases
  9. A one-line summary for your pitch

1What actually changes

Until now, AI agents could onlyread a store page's code and mimic clicks.
With WebMCP support, the store itself registers tools with the browser that say "call this" .

Before: read the code, mimic the clicks

The agent parses each page's DOM and imitates human actions. As the changelog puts it, this is slow and error-prone.

WebMCP: registered as tools

The page registers tools with the browser, and the agent simplycalls them by name. No guesswork.

Already live: enabled as of today on every Liquid storefront and in the Hydrogen Developer Preview.Nothing to install or configure(original: "There's nothing to install or configure.").

2How it works: traditional agents vs. WebMCP

BEFORE: read the page, mimic the clicks Agent Fetch HTML Parse the DOM Guess where the button is Structure differs by theme Simulate a click Re-parse on every page transition Slow / fragile slow and error-prone AFTER : the page registers tools with the browser Agent Tool call Browser (WebMCP) search_catalog update_cart Tools registered by the page Standard storefront actions The same path apps use Runs on the shopper's live session Theme behavior stays intact A theme that opens the drawer on cart update will open it for agent actions too
What is WebMCP : lets a page register tools with the browser —a proposed Web standard. Shopify is helping shape the spec alongside Google and Microsoft.
The key point: instead of "an agent poking at the UI on its own,"the site explicitly declares which actions may be called— that's the shift in direction.

3Tools an agent can call

Catalog
4 tools
Cart
3 tools
Checkout / Orders
2 tools
Policies / FAQ
1 tool

Catalog Catalog

search_catalogSearch the catalog
browse_storeBrowse the store
get_productGet a product
show_variantShow variants

Cart Cart

get_cartGet cart contents
update_cartUpdate the cart
cancel_cartCancel a cart

Checkout & orders Checkout & orders

proceed_to_checkoutProceed to checkout
manage_ordersManage orders

Other content Other content

search_shop_policies_and_faqsSearch shop policies and FAQs
The argument and return-value schemas, rate limits, and error behavior for each tool arenot documented in this announcement. Check the WebMCP documentation before you implement.

4Click emulation vs. WebMCP

AspectTraditional (click emulation)WebMCP tools
How actions are discovered Inferred reads the code on each page Declared the page registers its tools
Speed & reliability slow / error-prone(as worded in the original) No figures given in the article (it only states that improvement is the goal)
Where it runs Automated actions on the agent's side In the shopper's live session, inside the tab they are viewing
How cart actions are performed Reproduced by clicking the UI The same standard storefront actionsthat apps themselves use
Theme side effects — If the theme opens the drawer when the cart updates,it opens for agent actions too
Merchant effort — None no installation or setup required

5Availability and prerequisites

Liquid storefronts

All Liquid storefronts, available starting today.GA

Hydrogen

Developer Preview storefronts.preview

In the browser

Agent support is currently limited to an origin trial in Chromium-based browsers..

WebMCP is still a standard under development. Shopify is involved in shaping the spec alongside Google and Microsoft, butthe risk of spec changes remains..
The end date of the origin trial, the GA timing for Hydrogen, and the supported countries and plans arenot stated..

6What merchants need to do (3 steps)

1

No setup required

There is nothing to install and nothing to configure. On a Liquid storefront, it is already running.

2

Check your theme's behavior

If your theme opens a drawer or similar on cart updates, the same behavior fires for agent actions too.

3

Read the documentation

See the WebMCP documentation for details. If you're building your own agent, see "Build commerce agents."

75 points developers should know

1. It runs "inside the tab," not through a server-to-server API

Agent actions happenin the shopper's live session. The premise differs from external integrations that call the Storefront API behind the scenes: the login state, cart, and session are shared as-is.

2. Cart operations use "the same standard actions as apps"

Instead of a custom back door, it calls the same storefront actions existing apps use.Your existing cart hooks and event design keep working as-is..

3. Theme side effects fire for agents too

If your theme opens a drawer on a cart update, that drawer opens for agent actions as well.Modal and pop-up customizations need regression testing that assumes back-to-back operations driven by an agent..

TRIAL

4. Both the standard and the support are "in progress"

WebMCP is a proposed standard, and agent support is limited to an origin trial in Chromium-based browsers.Designing it as an additional channel rather than a primary production pathis the safer approach.

5. Read the tool names as a "public contract" of capabilities

The published tools are search_catalog / browse_store / get_product / show_variant / get_cart / update_cart / cancel_cart / proceed_to_checkout / manage_orders / search_shop_policies_and_faqs — 10 in total. The tools' scope ends at "proceed to checkout"; whether an agent carries the purchase through to completed payment is not stated. Because policies and FAQs are included in what gets searched,keeping your return policy and shipping terms pages in good shape directly affects the quality of an agent's answers..

83 use cases you can apply to your work

USE CASE 1

Take stock of whether your store can handle "visits via AI agents"

Challenge
On Liquid storefronts, the tools are already exposed with no setup required. That meansAgents can already come and operate the store without merchants doing anything. Custom theme behavior may fire in unexpected ways.
Action
In a test environment with a Chromium-based browser and the origin trial enabled, have an agent run straight through catalog search → variant display → cart update → checkout transition, and visually confirm how the theme behaves.
Impact
You can catch theme side effects like "updating the cart opens the drawer" that block consecutive agent actions, before they cause a real incident.
Technical notes
Cart operations go through the same standard storefront actions as existing apps, so your existing cart event monitoring works as an observation point as-is.
FAQ Returns Not set up
USE CASE 2

Rewrite policies and FAQs on the assumption that agents will read them

Problem
search_shop_policies_and_faqs is included in the public tools — meaning how you describe returns, shipping, and warranties directly determines the quality of the agent's answers. Vague FAQs produce vague answers.
Action
Return windows, shipping cost conditions, delivery times — the conditions that drive purchase decisions —spell them out explicitly, one question per answer: restructure your policy / FAQ pages into that shape.
Impact
Fewer drop-offs from "the conditions weren't clear" during agent-mediated purchase consideration. It also makes the pages easier to read for human visitors.
Technical notes
The granularity of what the tools return and the range of pages covered arenot documentedin the announcement. Hit it in a test environment to confirm what actually gets picked up.
search_catalog update_cart proceed_to_checkout
USE CASE 3

Move your own agent / customer-service bot onto the WebMCP tools

Problem
If your in-house shopping agent is built on DOM parsing and click simulation, it breaks every time the theme changes, and maintenance costs pile up.
Action
Shift to an implementation that calls the 10 published tools (4 catalog / 3 cart / 2 checkout and orders / 1 policy). The implementation guide is "Build commerce agents".
Impact
You can cut the dependency on theme structure. Because operations happen entirely within the shopper's live session, you also need less custom code for session syncing.
Technical notes
It only runs on Chromium-based browsers with the origin trial.A fallback path for unsupported browsers is essential. The standard is still being drafted, so wrap the tool-calling layer in a thin adapter to absorb spec changes.

9One-line summary for your pitch

"Stores now expose an operating surface for AI agents (10 WebMCP tools) on their own.
Liquid storefronts are already live with no setup required, Hydrogen is in Developer Preview, and the agent side is limited to the Chromium origin trial.
What to do now isn't implementation work — it's checking theme side effects and getting policies / FAQs in order."