WebMCP

Tools published by the page the assistant is embedded on.

Marked BROWSER in the connectors list, and enabled by default. WebMCP lets the page itself publish tools to the assistant, using the WebMCP standard.

WebMCP is a web standard, not a Maker one. Chrome's official WebMCP documentation covers how to declare tools on a page — read that if you're implementing your own.

Where other connectors are servers the assistant calls, WebMCP tools live in the browser, on the page the shopper is currently looking at — which means the assistant can act on the page, not just talk about it.

The canonical example: add to cart

A storefront can define an add-to-cart tool via WebMCP. A shopper viewing a product can tell the assistant "add this to my cart" — the assistant calls the page's tool and the item is actually added, on the page, with no custom integration work.

What else it can do

Add-to-cart is just the most obvious one. Because the tools run on the page the shopper is already on, WebMCP fits anything the shopper would otherwise do by clicking around themselves. A few examples:

• Change a product's options — the shopper says "show me this in medium, blue" and the assistant switches the variant on the page, updating the image and price.

• Update the cart — change a quantity, remove an item, or apply a promo code without the shopper leaving the conversation.

• Move through checkout — start checkout, or jump to the shipping or payment step, when the page exposes those as tools.

• Filter and sort a listing — on a category or search page, narrow results ("only show items under $50, in stock") by driving the page's own filters.

• Fill and submit a form — book an appointment, start a return, or sign up for restock alerts using the fields already on the page.

• Navigate the site — open the size guide, jump to reviews, or go to the shopper's orders, so the answer and the destination arrive together.

What's on offer is entirely up to the page: WebMCP lets a site expose any on-page action as a tool, and the assistant can use whichever ones it publishes. See Chrome's WebMCP documentation for how a page declares them.

Relationship to page context

WebMCP is the active counterpart to JavaScript API: context tells the assistant what is on the page, WebMCP lets it do something about it.

Like every connector, WebMCP sits in the priority order — a page-published tool can override a server-side one of the same name.

Access control

WebMCP tools are defined by the page at runtime, so there is no fixed list to configure in advance. You get the three group settings — Write tools, Read-only tools, Other tools — and a page-defined tool is offered to the assistant only if its group is set to Allow.

Disallowing Write tools here is the switch that stops the assistant acting on the page, add-to-cart included, while leaving it free to read page context. See Controlling tool access.