WebMCP: What it is, why it matters, and what to do now

Right now, many AI agents still interact with websites like a tourist navigating a foreign city without a map — and that’s the problem WebMCP is designed to solve.

They take screenshots, parse raw HTML, and guess what each button does. And if a pop-up suddenly appears, the agent gets disrupted and may click the wrong thing or start over.

It's slow, expensive, and unreliable.

WebMCP offers a shortcut, letting agents complete tasks without having to figure out every step of a website’s visual interface.

For marketers, this could fundamentally change website optimization. Alongside attracting and converting human visitors, you’ll also need to make your website work for AI agents.

In this guide, we'll cover what WebMCP is, how it works, whether it's ready to use, how it handles user consent, and what it means for SEO professionals and marketers preparing for the agentic web.

Shoutout to Vinicius Stanula at LOCOMOTIVE for inspiring this article!

What is WebMCP?

WebMCP (Web Model Context Protocol) is a proposed web standard that lets webpages make their features available to AI agents as structured, callable tools through a browser.

Developers can create these tools using HTML or JavaScript, allowing agents to perform specific actions such as searching for products or booking flights.

For example, say you asked an AI agent to find flights from Newark, New Jersey, to Austin, Texas, on Nov. 6. 

Here’s how that could work with and without WebMCP.

Without WebMCP: The AI agent has to figure out how to use the website. It may need to analyze screenshots, parse HTML, and work out where to enter information. Each step takes time and introduces opportunities for errors.

With WebMCP: The website makes a flight-search tool called “searchFlights” available to the agent. The tool specifies what information it needs, such as origin, destination, travel date, and number of passengers. The agent provides those details and gets a list of matching flights without navigating the site’s search filters or forms.

In other words, WebMCP turns your website into an API that AI agents can use, without you having to build or maintain a separate API.

WebMCP is still in development

WebMCP is still a draft specification, being developed in the World Wide Web Consortium (W3C) Web Machine Learning Community Group, with spec editors from Google and Microsoft.

It’s not an official W3C standard, but developers can already test it.

Google launched a public origin trial of WebMCP starting with Chrome 149, letting developers experiment with WebMCP on live websites.

How does WebMCP work?

WebMCP provides two ways for developers to make website features available to AI agents: a Declarative API and an Imperative API.

The Declarative API (HTML-based)

The Declarative API lets you turn existing HTML forms into tools that AI agents can use. 

If your website already has well-structured forms, you can make them accessible to AI agents by adding WebMCP attributes, including:

• toolname (required): Gives your tool a name that AI agents can identify and call
• tooldescription (required): Explains what the tool does, helping agents decide when to use it
• toolautosubmit (optional): Allows the form to submit automatically when an agent calls the tool
• toolparamdescription (optional): Provides additional instructions about what information a form field requires

For example, say you run a restaurant website with a reservation form.

You can add the two required attributes “toolname” and “tooldescription,” like this:

These two attributes signal to the browser that the form is an agent-callable tool. The browser then reads the form fields and automatically creates a JSON Schema that specifies what information the agent needs to provide, such as the reservation date, time, and party size.

The Imperative API (JavaScript-based)

The Imperative API is for more complex, dynamic interactions, where you need greater control over how AI agents interact with your website.

Unlike the Declarative API, which builds on HTML forms, the Imperative API lets you define tools directly in JavaScript.

Here’s what you need to do:

Register each tool through the browser’s document.modelContext interface, specifying four main properties:

• Name: Identifies the tool
• Description: Explains what the tool does
• Input schema: Defines what information the agent needs to provide
• Execute function: Defines the action your website performs when an AI agent calls the tool

For example, say you run a flight booking website. You could register a “searchFlights” tool that AI agents could use to search for available flights by departure airport, destination, and travel date.

Here’s a simplified example of how a flight-search tool could be registered using the Imperative API:

How AI agents use WebMCP tools

AI agents use WebMCP tools by calling them directly from an open tab. As soon as they navigate away or the tab closes, they lose access to those tools.

The process follows three steps:

With these three steps, an agent can replace multiple browser interactions — clicking filters, filling out forms, and analyzing screenshots — with one structured tool call.

This can reduce the time, tokens, and processing needed to complete a task. 

How does WebMCP handle user consent and security?

WebMCP supports user consent and security by letting websites expose specific actions to AI agents through the browser, rather than giving them unrestricted access to the site.

The browser manages how agents discover and call these tools. For sensitive actions, websites and agents can require user approval before proceeding.

Three mechanisms matter most for privacy and security:

• Manual submission by default: When an agent fills a form through the Declarative API, the user still submits it unless the site explicitly opts in to agent auto-submission with a “toolautosubmit” attribute
• User interaction requests: A tool's code can call “requestUserInteraction” to pause execution and ask the user to confirm a sensitive action, such as a purchase. This keeps a human in the loop before the action proceeds.
• Secure contexts and permission policies: The spec requires WebMCP to run on secure (HTTPS) pages and includes permissions policy controls

The security trade-off

WebMCP tools can let AI agents perform actions within a user’s logged-in sessions. This creates risks if an agent is manipulated into performing unauthorized actions.

For example, malicious instructions hidden in a product review or a tool response could trick an AI agent into revealing sensitive information — known as a prompt injection attack.

That’s why WebMCP includes security protections and ways to flag potentially risky actions, such as the “consequentialHint” annotation for sensitive actions.

But these mechanisms don’t guarantee an agent will behave safely.

So, your developers should also:

• Limit tools to specific, clearly defined actions
• Validate user permissions before allowing tools to perform actions
• Require confirmation before purchases, account changes, or deleting data

Is WebMCP production-ready? Current status and browser support

No, WebMCP is still an experimental proposal, not an official W3C standard.

But it is available for real-world testing in a public origin trial in Chrome and Edge.

Here's where each major browser stands:

*WebMCP status by browser, based on each browser vendor’s documentation and standards positions.

Should you implement WebMCP now?

You should implement WebMCP only as a small-scale test for now. It’s still too early to rebuild your website around it.

Here’s why you shouldn’t rush into full implementation:

• WebMCP is still evolving: The API change from “navigator.modelContext” to “document.modelContext” shows that some aspects of WebMCP are still being refined. Implementations you build today may need updating as WebMCP develops.
• Browser and agent support is limited: WebMCP remains experimental. ChatGPT’s desktop app now supports it through site tools, but only in the app’s built-in browser and for some accounts. Google has also announced plans to support WebMCP in Gemini in Chrome, and brands like Expedia, Booking.com, and Target are experimenting with it. But these are early adoption signs, not widespread support. 

Why should marketers and SEOs care?

Marketers and SEOs should care about WebMCP because it could change what makes your website successful.

As AI agents take on more tasks for users, agents could increasingly become the ones visiting your website and completing tasks.

This means how easily an agent can use your website could become a major factor in how much revenue you generate.

Three factors make it worth preparing for this shift.

Browsers are already preparing for an agentic web

Browser companies are already introducing features that signal a move toward an agentic web.

Testing WebMCP early could help you prepare your site before agent-driven browsing becomes more common.

Here’s how browsers are already making changes:

*Major agentic browser products and their launch dates.

WebMCP adoption could follow a similar path to responsive design

WebMCP could follow a similar adoption path to responsive design when mobile browsing took off.

Websites that adopted responsive design early were better prepared as more people started using the web on their phones.

WebMCP may still be experimental, with limited browser and agent support. But Google and Microsoft’s involvement in its development makes it worth tracking as agentic browsing continues to evolve.

Well-structured forms could lower WebMCP adoption costs

Well-structured HTML forms could make WebMCP easier and cheaper to implement, giving sites with accessible, clearly labeled forms a head start if the technology takes off.

The Declarative API builds on existing forms. So, the better structured your forms are, the less work your developers may need to do when you’re ready to adopt WebMCP.

Make sure your forms have:

• Clear descriptive labels
• Properly defined inputs
• Reliable form behavior

These improvements make your forms easier to use today while potentially reducing the development work needed to make them WebMCP-compatible later.

Real-world use cases

Here are some real-world use cases showing how WebMCP makes websites executable.

• Ecommerce: An agent could use “searchProducts,” “checkPrice,” and “addToCart” tools to find running shoes under $100 and add a pair to the cart for the shopper to review. The agent skips navigating product filters. For a deeper look at agentic commerce, see our guide to the Universal Commerce Protocol.
• Travel: A hotel booking site could let agents check available rooms and make reservations through “searchAvailability” and “bookReservation” tools. Google has a similar workflow in its WebMCP documentation. For a related example, see agentic restaurant booking in AI Mode.
• B2B SaaS: Tools like “createReport” and “exportData” in a sales dashboard could let an agent generate recurring reports and export data directly from a logged-in dashboard
• Customer support: An agent could call “createTicket” with the issue, category, priority, and description to submit a support request. It wouldn’t need to locate the support form or fill each field separately.

How WebMCP differs from MCP, OpenAPI, and llms.txt

WebMCP differs from Model Context Protocol (MCP) in where it runs: MCP connects AI systems to tools and data through a server, while WebMCP exposes a web page’s actions to an agent inside the browser.

*MCP and WebMCP are complementary — use both for full coverage

WebMCP vs. OpenAPI and llms.txt

OpenAPI and llms.txt are two other technologies you might encounter alongside WebMCP, but they serve different purposes.

• OpenAPI describes API functionality: OpenAPI describes the endpoints of a server-side API. WebMCP describes actions on a web page, and runs with the user's session in the browser.
• llms.txt points AI systems to web content: llms.txt is a file that points AI systems to the content on your site you want them to read. WebMCP covers what an agent can do on your site, not what it can read. See our llms.txt guide for more.

How to test WebMCP today

To test WebMCP today, use Chrome 149 or later (or Edge 150 or later). Then, either join the origin trial or turn on the local testing flag. The origin trial works best for sites that want to test with real users, and the flag is the fastest way to experiment on your own machine.

Option 1: Join the origin trial

Here’s how to join the origin trial.

1. Register your domain: Sign up for the WebMCP origin trial through Chrome for Developers or the Microsoft Edge Developer Center
2. Add your trial token: Add your origin trial token to your site’s HTTP response header or as a <meta> tag inside your HTML <head>
3. Enable automatic access: Visitors using supported Chrome or Edge versions can now interact with your WebMCP tools

Option 2: Test locally with the flag

Testing locally with the developer flag is ideal when you’re building tools in local dev environments or experimenting without needing a registered domain.

Follow these steps:

Step 1: Verify your browser version. Make sure you’re running Chrome 149 or later, or Edge 150 or later. Verify at “chrome://settings/help” in Chrome or “edge://settings/help” in Edge.

Step 2: Enable the WebMCP testing flag. Navigate to “chrome://flags/#enable-webmcp-testing” or “edge://flags/#enable-webmcp-testing” in Microsoft Edge. Then change the dropdown setting from Default to “Enabled.”

Step 3: Relaunch your browser. Click the “Relaunch” button at the bottom of the page.

Step 4: Install the Model Context Tool Inspector extension. Download it from the Chrome Web Store. It lets you inspect registered tools on any page and test them with custom parameters.

Step 5: Test on live demos or local web apps. Test your local setup using your own web app or Google’s open-source demos such as the live travel demo.

What WebMCP means for AI visibility

WebMCP adds a new dimension to AI visibility: whether AI agents can act on your brand, not just mention or recommend it.

Until now, AI visibility has largely focused on getting your brand cited and recommended in AI-generated answers.

But a recommendation doesn’t guarantee a conversion.

For example, say an AI recommends your hotel. But if the agent struggles to check room availability, compare prices, or complete the booking, that recommendation might not turn into revenue.

WebMCP could help close that gap.

The visibility stack is expanding

The AI visibility stack has a new layer — agent readiness.

The three layers reinforce one another.

SEO helps your business get discovered. AI visibility helps put your brand in relevant AI-generated answers. And agent readiness makes it easier for compatible AI agents to complete tasks once they reach your website.

You don’t have to wait for WebMCP to take off to start preparing. You can already strengthen the first two layers: SEO and AI visibility.

Start by seeing where you stand today.

Check your SEO performance and AI visibility to see what’s working and what you need to improve.

You can do all that with Semrush One. It’s an all-in-one platform that brings together the tools you need to track and improve your visibility in search engines and AI search.

What to do right now

What you can do right now is prepare your website for the agentic web.

WebMCP is still experimental, but much of the groundwork overlaps with good technical SEO: well-structured HTML, accessible forms, and reliable website functionality.

These improvements benefit your website today and could make WebMCP easier to adopt.

Here are three ways to do that.

Get your AI visibility baseline

Get your AI visibility baseline now so you can track how your brand’s presence in AI search changes as the agentic web develops.

Start with the Semrush AI Visibility Toolkit to see where your brand is already visible.

Check:

• Visibility Overview: Get your domain’s AI Visibility Score and see how often your brand appears in AI answers
• Competitor Research: Find the topics and prompts that competitors show up for, but your brand doesn’t
• Brand Performance:Track brand sentiment and how AI talks about your brand

For larger teams managing multiple brands or markets, Semrush Enterprise AIO offers AI visibility reporting on ChatGPT, Perplexity, Gemini, and other AI search platforms. It also provides executive-ready dashboards, so it’s easier to report results to leadership as WebMCP evolves.

Audit your key user actions

Audit your key user actions to identify which actions could benefit from WebMCP and what might prevent AI agents from completing them reliably.

Start with the five to 10 most important actions on your site, such as:

• Submitting lead forms
• Searching products
• Making bookings
• Completing checkout
• Requesting support

For each action, check that:

• Form labels are clear and descriptive
• Input fields are properly defined
• Forms validate inputs and handle errors
• Pages and submission flows work reliably

Then check those pages’ technical health at scale. 

Run a Semrush Site Audit to identify broken links, redirect problems, and crawlability issues. Also check your robots.txt to confirm AI systems can reach those pages.

For a WebMCP-specific check, run Lighthouse’s experimental Agentic Browsing audit. It checks how well your pages work for AI agents, including whether any WebMCP tools you’ve added are registered correctly.

For more on keeping your site healthy for AI search, see our guide to site health and AI search visibility.

Start the conversation with your team, clients, and stakeholders

Start conversations about WebMCP with your team, clients, and stakeholders so they understand how the agentic web could affect your website and business.

Frame these conversations as awareness, not a push for immediate implementation. Include what's coming, why it matters, and how the work you’re already doing positions the business well.

Here’s where to start:

• Talk to your developers: Share this article and point them to the Chrome for Developers WebMCP documentation and the W3C Web Machine Learning Community Group
• Introduce WebMCP to clients: If you work at an agency or as a consultant, explain how AI agents could change the way customers interact with their websites. Focus on potential opportunities without recommending a major investment yet.
• Brief internal stakeholders:Help decision-makers understand where WebMCP fits into your broader AI search strategy, what’s still uncertain, and why technical SEO investments could make adoption easier

Further reading: How to optimize for the agentic web: a guide for marketers

WebMCP FAQ

What is WebMCP?

WebMCP (Web Model Context Protocol) is a proposed browser standard that lets websites expose actions — such as searching, booking, or submitting forms — as structured tools that AI agents can call.

Instead of forcing AI agents to inspect webpages and simulate clicks, WebMCP gives them clearly defined functions they can execute. It replaces much of the guesswork involved in screen-scraping and browser automation with structured, predictable interactions.

How do I enable WebMCP in Chrome?

To enable WebMCP in Chrome, use Chrome 149 or later, and go to chrome://flags/#enable-webmcp-testing. Find the “WebMCP for testing” setting and change it to "Enabled." Then relaunch the browser to activate. Sites can also join the WebMCP origin trial so users get it without changing browser settings.

Does WebMCP replace MCP?

No, WebMCP doesn’t replace MCP. They’re complementary technologies, and you can use both. WebMCP enables agents to handle in-browser interactions on a web page, while MCP handles server-side tools and data access.

Do I need WebMCP for SEO or AI visibility today?

No, you don’t need WebMCP to appear in AI-generated answers. It’s still experimental, but you can start preparing now. Make forms easier to use, simplify user journeys, and ensure your website works reliably. These improvements benefit visitors today and make it easier for AI agents to interact with your site later.

Can I add WebMCP to a WordPress form?

Yes, as long as the form uses standard HTML and you can modify its attributes. Add WebMCP attributes, such as toolname and tooldescription, to the form’s HTML. You can do this manually or through a form plugin that supports custom attributes. To check that Chrome recognizes the tool and that it works correctly, use the Model Context Tool Inspector Chrome extension.

Is WebMCP safe?

WebMCP includes safeguards, but it isn't risk-free. AI agents can be vulnerable to prompt injection and may perform unauthorized actions. Reduce these risks by limiting what each tool can do, testing implementations in a sandbox, and requiring user confirmation for sensitive actions such as purchases or account changes.

WebMCP resources

For more information on WebMCP, here are some resources:

• WebMCP on Chrome for Developers: Documentation and the origin trial
• WebMCP draft specification: The W3C Web Machine Learning Community Group draft
• Model Context Tool Inspector: Chrome extension for inspecting and testing registered tools
• Google's WebMCP flight search demo: A live example of tool discovery and execution

WebMCP and getting ready for the agentic web

WebMCP offers a promising way for AI agents to interact with websites, without the fragility of screen scraping or the overhead of maintaining separate APIs. 

But it's still an origin trial. Plus, few agents use it yet, so focus on what you can control.

Measure your AI visibility. Clean up the forms and flows of your most valuable website actions, and bring your developers into the conversation.

These improvements already benefit your website today. And if WebMCP becomes widely adopted, you’ll have a stronger foundation to build on. For more on how to measure and report on AI search visibility.