- Published
- Updated
- 11 min read

TL;DR: WebMCP is a proposed web standard, edited by engineers at Google and Microsoft, that lets a site hand AI agents a short list of tools instead of making them guess at buttons. It is in a Chrome origin trial, not finished. An agent-ready website starts with clean forms, true pages and clear actions; WebMCP comes after that.
An agent-ready website is one an AI agent can use on a person's behalf without guessing. Today most sites are built for one audience: a human with a mouse or a thumb. When an agent shows up, it reads the page, looks at the layout, and tries to work out which button books the appointment.
Google and Microsoft have proposed a way to skip the guessing. It is called WebMCP, and it lets a page tell an agent, in plain terms, what it can do there. I want to walk through what actually shipped, what the numbers you may have seen really measure, and what a small business should do with its site this year. Some details in the first version of this note were wrong, and I have corrected them below against the specification and Chrome's own documentation.
What is WebMCP, in plain words?
WebMCP lets a web page describe its own actions as tools. The project's explainer puts it this way: it "lets developers expose web application functionality as 'tools'", described as "JavaScript functions with natural language descriptions and structured schemas, designed for AI agent ingestion." (WebMCP explainer, Web Machine Learning Community Group (opens in a new tab))
The name borrows from the Model Context Protocol (MCP), which AI platforms already use to call outside services. The explainer describes a page that uses WebMCP as something like an in-page MCP server, with tools that run in the browser instead of on a separate backend. For an agent-ready website, that is the whole idea: say what the page does instead of hoping the agent works it out.
Chrome's documentation explains the problem it solves. Without it, an agent relies on actuation, which Google defines as "the act of an agent simulating manual mouse clicks and text input, as though it were the human user engaging with your website." With WebMCP, "the website declares the element's purpose, so it's used as intended." Google calls this "more reliable than actuation, which may have numerous steps and leaves each step open to interpretation by the agent." (Chrome for Developers (opens in a new tab))
What does an agent-ready website look like with WebMCP?
There are two ways to add tools, according to Chrome's documentation. (Chrome for Developers (opens in a new tab))
- The imperative API. A developer writes a little JavaScript that registers a tool with the page: a name, a description, a schema for the inputs, and the function that runs. The specification places this on a new model context object on the page's document, with a method for registering tools. (WebMCP specification (opens in a new tab))
- The declarative API. A developer adds annotations to an existing HTML form, so the browser can turn that form into a tool. The draft explainer for this approach uses attributes named toolname and tooldescription on the form element. (WebMCP declarative API explainer (opens in a new tab))
One correction matters here. Our first version showed a sample form with made-up attribute names. The real proposal uses the names above, and the main explainer says the declarative approach "is being explored separately" and "is not part of the specification at this time." (WebMCP explainer (opens in a new tab)) For an agent-ready website, that means the form-only shortcut may change before it settles.
Where does WebMCP stand in October 2026?
It is real, it is moving, and it is not a finished standard. Here is the timeline from primary sources, and what each step means for an agent-ready website.
- February 10, 2026: Google opened WebMCP to its early preview program for prototyping. (Chrome for Developers (opens in a new tab))
- June 9, 2026: Google announced that "In Chrome 149, you can sign up for the WebMCP origin trial." (Chrome for Developers (opens in a new tab))
- Edge: the project's status page lists an origin trial "live in Edge 150." Firefox and Safari are listed only with open standards position threads, not an implementation. (WebMCP implementation status (opens in a new tab))
The specification is a Draft Community Group Report from the W3C's Web Machine Learning Community Group, with editors from Microsoft and Google. Its own status section says it "is not a W3C Standard nor is it on the W3C Standards Track." (WebMCP specification (opens in a new tab))
What does an origin trial mean for an agent-ready website?
Google describes origin trials as "time-limited programs that offer early access to experimental platform features," with possible usage limits. (Chrome for Developers (opens in a new tab)) Chrome's documentation also says WebMCP "is under active discussion and subject to change in the future." (Chrome for Developers (opens in a new tab))
So treat it like any early web feature. Learning it is cheap. Rebuilding your whole site around it today is not necessary. Nobody I know of is losing customers this month because their agent-ready website lacks WebMCP tools.
Do agents really work better with structured tools?
The reasoning is sound. An agent that calls a named tool with a defined schema has fewer chances to misread a page than one clicking through it screen by screen. That is the case for an agent-ready website in one sentence. Chrome lists three things WebMCP supports: discovery of tools, JSON Schemas that define inputs and outputs "to reduce hallucination or misunderstanding," and shared page state. (Chrome for Developers (opens in a new tab))
The cost figures that circulated with this news need a second look. The first version of this note quoted a paper reporting $0.0051 per call against $0.0110. That paper is a single-author preprint about a different proposal with a similar name: "webMCP (Web Machine Context & Procedure)," which "embeds structured interaction metadata directly into web pages." It reports an evaluation across 1,890 real API calls and a 67.6% reduction in processing. (arXiv (opens in a new tab)) Those results describe that design, not the Google and Microsoft proposal, so we removed them. Chrome's WebMCP documentation does not publish cost figures.
What limits does Google list for WebMCP?
Chrome's documentation names three. (Chrome for Developers (opens in a new tab))
- Headless browsing: the API "is primarily designed for local browser workflows with a human in the loop."
- Complex interfaces: a complicated site will "likely need to refactor or add JavaScript to handle application and interface state."
- Discoverability: "Clients and browsers must visit a site directly to know if it has callable tools."
That last point is worth sitting with. WebMCP helps an agent once it is on your page. It does nothing to get the agent there. For an agent-ready website, being found still comes first, and that is where most sites I review fall short.
Is WebMCP the only way to build an agent-ready website?
No. Several efforts are aimed at the same audience, and they solve different problems.
- llms.txt is a proposal, first published by Jeremy Howard on September 3, 2024, "to standardise on using an /llms.txt file to provide information to help agents use a website." (llmstxt.org (opens in a new tab)) It is a text file for reading, not a way to take actions.
- The Universal Commerce Protocol is Google's "open-source standard designed to power the next generation of agentic commerce," aimed at checkout and shopping. (Google Developers Blog (opens in a new tab))
- The Agentic Commerce Protocol describes itself as "An open standard for programmatic commerce flows between buyers, AI agents, and businesses," released under the Apache 2.0 license. (Agentic Commerce Protocol (opens in a new tab))
WebMCP's difference is that the browser itself hosts the tools on the page a person is already using. The commerce protocols matter most if you sell products online. For most service businesses planning an agent-ready website, the browser route is the one to watch.
What does Google say you need for AI search features?
Less than many vendors suggest. Google's guidance for site owners says "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." It also says "You don't need to create new machine readable files, AI text files, or markup to appear in these features." (Google Search Central (opens in a new tab))
The same page asks site owners to make sure structured data "matches the visible text on the page." That is the quiet foundation of an agent-ready website: what a machine reads should be what a person sees. I wrote more about how assistants pick sources in how AI assistants decide which businesses to recommend (opens in a new tab).
What should a small business do with its website this year?
Think in layers. Being found in search comes first. Being quoted in AI answers comes second. Letting an agent take an action comes third, and that is where WebMCP lives. Most business sites still have work to do on the first two, so start there.
- Fix your forms. Every field gets a visible label, a clear purpose and a working error message. A form a person can finish without help is the base of an agent-ready website, with or without WebMCP.
- Write true, specific pages. Answer the questions your customers ask, in text, with facts you can back.
- Keep structured data honest. Mark up your business details and services, and make sure the markup says what the page says.
- List your actions. Write down the three to five things a visitor actually does on your site: book, request a quote, search, subscribe, pay. Those are your future tools.
- Decide what an agent may never do alone. Purchases, cancellations and account changes should ask the person first. Chrome's documentation notes you "can include a command to request user interaction with a confirmation dialog" for sensitive actions. (Chrome for Developers (opens in a new tab))
- Plan WebMCP into your next rebuild. If a redesign is coming in the next year, ask for the actions above to be built so they can become tools later. Developers can test today with a Chrome flag for local development.
Which sites gain the most from an agent-ready website first?
Sites with clear transactions. Google's own examples are customer support, ecommerce and travel booking (Chrome for Developers (opens in a new tab)), plus filling in "complex, structured forms." (Chrome for Developers (opens in a new tab)) A clinic intake form, a quote request for a contractor or a reservation flow at a restaurant fits that pattern. A purely informational site gets most of its benefit from clean pages and good structured data.
What security questions come with an agent-ready website?
Real ones. The WebMCP specification's security section covers prompt injection, including "Tool Poisoning" through tool descriptions, and "Misrepresentation of Intent," where a tool does something other than what the agent expects. (WebMCP specification (opens in a new tab)) Chrome adds that both APIs are gated by a tools permissions policy that, by default, blocks tool registration in cross-origin iframes. (Chrome for Developers (opens in a new tab))
If you add tools, treat each one like a public API. Validate every input, keep descriptions short and accurate, and log what runs. Our note on agentic security covers the wider picture.
What did the courts say about AI agents acting on a website?
The law is still forming, and the main case moved during 2026. Amazon sued Perplexity in November 2025 over its Comet browser's shopping agent. In March 2026 a federal judge granted a preliminary injunction, writing that Amazon had provided "strong evidence" that Comet accessed its site at the user's direction but "without authorization." (CNBC (opens in a new tab))
On August 4, 2026, the Ninth Circuit vacated that injunction. The panel wrote: "It is the user who 'accesses' Amazon's computers, with the help of the Assistant to carry out specific acts on Amazon.com." It added that the outcome "does not impair Amazon's ability to regulate access to Amazon.com via private terms of service for its users." (U.S. Court of Appeals for the Ninth Circuit (opens in a new tab)) The case went back to the district court.
Our first version said this ruling would push agents toward sites that declare their tools. The appeal changed that picture, so I will say only this: an agent-ready website that publishes WebMCP tools is stating which actions it intends agents to take, and your terms of service still matter. Ask your attorney before you rely on either.
Frequently asked questions
What is WebMCP?
A proposed web standard that lets a page register tools, each with a name, description and input schema, so an AI agent can call them instead of clicking through the page. It is a W3C community group draft, not a W3C standard.
Can my customers use WebMCP today?
Only in limited ways. Chrome runs an origin trial from version 149 and Edge from version 150, and developers can test with a flag. Firefox and Safari have not shipped it.
Do I need an llms.txt file to appear in Google's AI answers?
No. Google says you do not need new machine readable files or AI text files to appear in AI Overviews or AI Mode. Normal SEO practice still applies.
What makes an agent-ready website?
Clear forms with labels, true and specific pages, structured data that matches the visible text, and a short list of actions a visitor takes. WebMCP tools can be added on top of that later.
Is it safe to let AI agents take actions on my site?
It can be, with care. Treat each tool like a public API, validate inputs, ask the person to confirm purchases or account changes, and watch for prompt injection through tool descriptions.
Should I rebuild my site for WebMCP now?
Usually not. If a rebuild is already planned in the next year, design your main actions so they can become tools, and an agent-ready website becomes a small step later. If the site works and traffic is steady, keep watching the standard.
This article is general information, not legal advice. Kief Studio is not a law firm, and reading it does not create an attorney-client relationship. Laws and regulations change, so check the current text and talk with a licensed Massachusetts attorney about your situation.
Where we fit
Kief Studio builds custom web applications and public sites around how a business already works. We map how the work happens today before choosing a screen or a stack, build the forms and flows people actually use, and host them on dedicated infrastructure. That same map of actions is what an agent-ready website needs when WebMCP settles.
On the search side, our SEO, AEO and GEO work finds the questions your buyers ask, writes pages that answer them with facts you can back, and adds structured data so search engines and assistants can read each page. If keeping the site current has become a second job, managed website operations covers hosting, publishing and security in one place.
First conversation is free. No commitment.




