What is WebMCP? Your contact form is about to become an agent's tool

The short answer

We have argued that what an agent can find, read and act on is becoming the conversion layer. Our guide to blocking or allowing AI agents covers whether to let them in. WebMCP is the next question: once an agent is on your page, can it do the job it came for?

What is WebMCP?

WebMCP is a browser feature that lets a web page publish a list of actions, with a name, a description and the inputs each one needs, for an AI agent working in that browser. The agent reads the list and calls the action it needs. It does not have to work out which field is which or where the submit button is.

The draft specification puts it in one line: the API "enables web applications to provide JavaScript-based tools to AI agents" (W3C Web Machine Learning Community Group). It is published as a Draft Community Group Report, most recently dated 27 October 2026. That is a proposal under discussion, not a ratified standard.

Google describes it as "a proposed web standard to help you build and expose structured tools for AI agents" that is "under active discussion and subject to change" (Chrome for Developers). When Chrome opened an early preview in February 2026, it said the aim was agent workflows that are more reliable and faster than "raw DOM actuation", meaning an agent clicking and typing its way through a page as a person would (Chrome for Developers).

The name borrows from the Model Context Protocol (MCP), the standard many AI assistants already use to connect to outside tools. The difference is where the tools live. A normal MCP server runs on a company's servers and an assistant connects to it over the internet. WebMCP tools live inside the web page, run in the visitor's browser, and work in the session the person already has open.

How does WebMCP work on a website?

There are two ways to add it. The declarative way adds attributes to an HTML form you already have. The imperative way registers tools in JavaScript for actions that are not a simple form, such as filtering a product list or updating a cart.

The declarative form is the one most marketing teams will meet. Chrome's documentation lists four attributes (Chrome for Developers, declarative API, updated 25 September 2026):

• toolname on the form names the tool, for example "request_quote".
• tooldescription on the form says what the tool does and when to use it.
• toolparamdescription on a field describes that input, for example "postcode where the work is needed".
• toolautosubmit is optional. Without it, "the user must manually click Submit to complete the task".

That last default matters. An agent can fill your enquiry form through WebMCP, but unless you opt in to automatic submission, the person sees the completed form and presses the button. The human stays in the loop for the step that creates the lead.

The imperative form uses document.modelContext and its registerTool() method in the draft specification. Older tutorials show navigator.modelContext; the specification now places the API on the document, so check the date on any guide you follow. Chrome gates both forms behind a permissions policy that allows tools on the page itself and same-origin frames, and blocks cross-origin iframes unless they are explicitly allowed. It also suggests a confirmation step for sensitive actions "such as making a purchase".

Who has switched WebMCP on so far?

Ecommerce platforms first, because a store's actions are standard: search, view a product, add to cart, check out. Service businesses and lead-generation sites have no platform doing it for them yet.

Shopify. The developer changelog added WebMCP for Liquid and Hydrogen storefronts on 5 August 2026 and for checkout on 28 September 2026 (Shopify changelog). The storefront tools cover searching the catalogue, browsing, viewing products and variants, managing the cart, proceeding to checkout, managing orders and searching the shop's policies and FAQs. Shopify's checkout documentation tells agent builders to "show the buyer the current order and total, and get their permission to place it" before the order is placed (Shopify). Our guide to Shopify agentic commerce explains how this sits beside Shopify's server-side catalogue and checkout for agents.

Commerce.com and Cloudflare. Cloudflare injects a small script at the edge that exposes Commerce.com's shopping tools, such as search_products and add_to_cart, on each storefront. Merchants can opt out from their dashboard. If the store's MCP server is offline, or the visitor's browser does not support WebMCP, the script exits and the page loads as normal. The shopper still clicks checkout to finish the purchase (Cloudflare).

Who calls these tools? Fewer agents than the rollout numbers suggest. Cloudflare's post calls WebMCP "an experimental proposed browser standard in early preview" and lists Chrome flags, Chrome origin trials, "some agentic browsers" and developer environments as the ways to use it today. Shopify says agent support "is currently limited to Chromium-based browsers" (Shopify). Neither names a mainstream consumer assistant that calls the tools today, so treat this as infrastructure being laid ahead of the traffic.

How is WebMCP different from MCP, llms.txt and screen-reading agents?

Each one answers a different question an agent has about your website. llms.txt says what the site is about. A server-side MCP connection gives an assistant your data and actions without loading the page. WebMCP describes what the page in front of the agent can do. Screen reading is the fallback when none of those exist.

The rows are layers, not rivals. Commerce.com runs WebMCP alongside its server-side MCP, using the same shopping tools for both. And the last row never goes away: an agent that lands on a page without WebMCP still has to read it. Our llms.txt guide explains why the signpost file is cheap to add and no substitute for a page that works.

Does a marketing website need WebMCP?

Not urgently, but sooner than most teams expect. A service business with an enquiry form gains little from WebMCP until assistants that use it reach its customers. The cost of the declarative version is small, though, and Google has already built a check for it.

Chrome's Lighthouse has an experimental Agentic Browsing category with three WebMCP audits: registered tools, schema validity, and "Forms missing declarative WebMCP" (Chrome for Developers). It gives no score out of 100, because the standards are still settling. But a Google tool that flags an ordinary contact form for lacking agent markup tells you where the browser is going.

That line, from Lighthouse's scoring notes, is the more urgent point. The same audits check accessibility for agents and layout stability. A form whose fields have no proper labels, whose error messages appear only as red borders, or whose submit button jumps as the page loads will fail an agent with or without WebMCP. Our contact form best practices cover the fields and errors that lose human enquiries, and the same fixes help agents.

On WordPress, the enquiry form usually comes from a form plugin, so adding WebMCP attributes means a plugin setting, a filter or a small template change, tested in Chrome with the WebMCP flag turned on. That is routine delivery work for a WordPress specialist such as WP Creative, and it is worth doing in the same pass as fixing the form's labels and errors rather than as a separate project.

Where is WebMCP heading?

Our view: WebMCP is the most important agent standard most marketers have never heard of, because it moves the conversion point. Today an agent has to read a page the way a person does and hope it guesses right. With WebMCP, the page tells the agent what it can do. The enquiry form stops being a layout and becomes an interface.

We expect three things over the next 18 months. These are forecasts, not facts.

• Platforms will do it for you. Shopify and Commerce.com did not ask merchants to write code. We expect form builders, booking systems and CMS platforms to add WebMCP attributes to forms by default, the way they added spam protection. Most small businesses will inherit WebMCP rather than build it.
• Human confirmation will stay the default for anything that matters. Chrome's declarative default waits for the person to press submit. Shopify asks agents to get the buyer's permission before placing an order, and Commerce.com leaves checkout to the shopper. Agents will fill in more forms. People will keep pressing the final button for some time yet.
• Browser support will decide the pace. WebMCP works in Chromium browsers behind an origin trial. Until it is on by default in Chrome and has a second browser engine behind it, it is an advantage for early movers, not a requirement.

The risk for business websites is not missing WebMCP this quarter. It is building forms and pages that only work with a mouse, a script and a patient human, and finding that agents quietly take the enquiry elsewhere.

What to do next

1. Run the agent readiness check to see whether an agent can read your pages and find your contact path without running scripts.
2. Run can visitors contact you and fix any form fields without visible labels, unclear errors or a CAPTCHA before the form loads.
3. Run Lighthouse's Agentic Browsing category in Chrome 150 or later on your main enquiry page, and read the accessibility and layout findings first.
4. Ask your platform what it plans. If you run Shopify or Commerce.com, WebMCP is likely already on: check that your product data and policies are complete, because agents will read them. Otherwise, ask your form plugin or website team whether WebMCP is on their roadmap.
5. Prototype one form. Add toolname, tooldescription and toolparamdescription to your main enquiry form, leave automatic submission off, and test it with Chrome's WebMCP flag before it goes live.
6. Keep the person in charge of the final step. Leave the submit, payment and account changes to a human confirmation.

Frequently asked questions

What is WebMCP?

WebMCP is a proposed web standard that lets a web page tell an AI agent in the browser which actions it offers, such as searching products or sending an enquiry, and what inputs each one needs. The agent calls those tools directly instead of clicking through the page. It is a W3C Community Group draft, in a Chrome origin trial.

What is the difference between WebMCP and MCP?

An MCP server runs on a company's servers and an AI assistant connects to it over the internet, with no web page involved. WebMCP tools live inside a web page and run in the visitor's browser, in the session the person already has open. A business can offer both, as Commerce.com does.

Which browsers support WebMCP?

Only Chromium-based browsers, and only experimentally. Chrome offers WebMCP through an origin trial from Chrome 149 and through a testing flag. Shopify says agent support is currently limited to Chromium-based browsers. Firefox and Safari have not shipped it.

Do I need WebMCP for SEO or AI search?

No. WebMCP helps an agent act on a page it has already reached, not find or cite your page. Search visibility and AI citations still depend on crawl access, readable server-rendered pages and clear facts. Lighthouse checks WebMCP in an experimental agent category that has no score out of 100.