AI NewsModels & agentsAnnouncement
WebMCP goes live at Shopify, Cloudflare and OpenAI for browser agents
Shopify, Cloudflare and OpenAI have put WebMCP into live products in the same three-week window, letting a browser agent call named tools on a page instead of imitating clicks.

Image: Search Engine Journal
Why it mattersA page can now hand an agent a typed action with a description and a schema, so a checkout, a search or a filter runs as one structured call that carries defined inputs and returns defined data.
Three companies added WebMCP to released products inside three weeks. Shopify enabled a set of WebMCP tools on every Liquid and Hydrogen storefront on 5 August. Cloudflare released a developer preview on 6 August that adds a WebMCP bridge at the edge without touching the origin. OpenAI added Site tools to ChatGPT's built-in desktop browser on 25 August, so ChatGPT Work and Codex can now discover and call tools that a page provides.
Google Chrome opened an origin trial for the proposal in Chrome 149 in June, and Search Engine Journal reports that OpenAI has since started a ten-day WebMCP Challenge with Chrome, Shopify, Cloudflare, Netlify, Vercel and Render as backers.
What WebMCP changes
A browser agent that works today usually reads a screenshot or a DOM tree, decides what a button does, and imitates a click. Small changes break the process: a renamed label, an unexpected pop-up, a custom date picker.
WebMCP works differently. A page registers a named tool with the browser, with a description and a structured input schema. The agent asks the browser what tools this page exposes, picks one, sends the required fields, and receives structured data back. Chrome's own comparison of WebMCP with MCP puts it this way, quoted by Search Engine Journal: "Instead of your application being a guest within an agent, the agent is a guest on your platform."
What each product includes
Shopify's changelog says it publishes ten WebMCP tools on every Liquid storefront, and the same set is available on Hydrogen storefronts in the developer preview. One of them, proceed_to_checkout, moves the shopper into checkout with the current cart, but does not complete the purchase. Cart updates happen inside the customer's active tab, and Shopify's platform picks the initial tool set for every hosted store. Shopify's WebMCP documentation does not say whether merchants can turn individual tools off or edit their descriptions.
Cloudflare's preview adds a single line at the edge to each HTML response, pointing at a same-origin bridge script that checks browser support and registers the selected tool packs. In this preview the tools execute inside the visitor's browser, so an authenticated session stays local to the tab and does not cross into Cloudflare infrastructure.
Who sets the defaults
The interesting part for engineering teams is control, and who has it. WebMCP gives a website the ability to guide an agent's behaviour, but on a hosted Shopify store the platform picks the default tools, and on a site behind Cloudflare the infrastructure provider is the party inserting the bridge. If your product depends on how an agent interacts with a hosted storefront, the tools the agent sees may start with defaults set by another company, before your merchant preferences apply.
The security risks also change. Because a WebMCP tool call runs inside the visitor's authenticated tab, it inherits cookies, session state and page context. Chrome's own guidance points out two current risks: tool definitions that hide instructions inside a name or description, and responses that carry embedded instructions from third-party data. Prompt injection does not go away because the surface is structured; it takes a different form.
Source
- Shopify changelog: WebMCP support for Liquid and Hydrogen storefronts
- Cloudflare blog: Give any website a WebMCP interface
- Search Engine Journal: WebMCP Connects AI Agents To Actions Inside Websites
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.

