What is WebMCP? How it works and why it matters for AI agents

AI agents can search websites, fill out forms, and complete multi-step tasks on your behalf. To do that, they often need to interpret an interface built for people: find the right fields, work out which buttons to click, and navigate between pages.

That leaves room for errors. A redesigned page or an unfamiliar form label can make a task harder for an agent to complete.

WebMCP lets websites describe their available actions as structured tools. An agent helping you book a flight, for example, could use a tool for searching flights instead of working through the search interface. Developers define what the tool does and which inputs it accepts, giving agents a more predictable way to use the website.

What is WebMCP?

WebMCP is a proposed web standard that lets websites expose structured tools to AI agents operating in a browser. Agents can discover and use these tools without having to interpret the visual interface or simulate clicks and keyboard input for each action.

It takes inspiration from the Model Context Protocol (MCP), which standardizes how AI applications interact with external tools and data sources. WebMCP applies that idea within the browser, where an agent and a user can work with the same open website.

Consider an agent helping you book a flight. Without WebMCP, it might need to identify the booking fields, enter your details, and click through multiple screens. With WebMCP, the website could expose tools that describe how to search for flights or complete a booking.

The WebMCP specification is being developed through the W3C Web Machine Learning Community Group, with contributions from engineers at Google and Microsoft. It’s a draft Community Group Report, not a finalized W3C standard or a specification on the W3C Standards Track.

How does WebMCP work?

Browser-based agents typically inspect HTML or use screenshots to understand a page. They identify controls, decide which action to take, and interact with the interface. A moved button or an ambiguous label can interrupt that process.

WebMCP gives developers a way to describe those actions explicitly:

1. The website exposes tools for actions such as searching a product catalog, submitting a form, or retrieving information.

2. A compatible browser makes the tools available to an AI agent, along with descriptions and the inputs they accept.

3. The agent selects and invokes a tool in response to the user’s request, using the website’s existing functionality.

An online store, for example, might expose a tool called search_products that accepts a product category, price range, and keyword. An agent could call it with the relevant parameters and receive structured results, saving it from having to navigate the store’s search controls.

The website itself defines how each tool works, so developers have more control over how agents use its functionality.

Two ways to implement WebMCP

WebMCP supports two approaches to exposing tools:

• The declarative API lets developers annotate existing HTML forms so agents can understand and interact with them. It suits simpler actions that already have a form on the website.

• The imperative API lets developers register tools using JavaScript, with more control over complex interactions and application functionality.

Both approaches make existing website functionality available to AI agents without requiring a completely separate interface. Google’s WebMCP documentation explains how to use them.

WebMCP vs MCP: what’s the difference?

WebMCP and MCP both help AI agents interact with external systems through defined tools. Their main difference is where that interaction happens.

MCP connects AI applications to external tools, services, and data sources. An AI coding assistant might use an MCP server to search documentation, access a database, or interact with a development platform.

WebMCP focuses on the website a user has open in their browser. The website exposes tools that a compatible browser agent can discover and use in that context.

WebMCP complements MCP, rather than replacing it. A business could use an MCP server to make its services available to AI applications outside the browser, then use WebMCP to support agents working directly on its website. Supporting both gives people more ways to access those services through their preferred AI tools.

What can you do with WebMCP?

Web applications can expose tools for tasks that users might ask an AI agent to perform. The tools depend on the website’s functionality, from finding a product to retrieving a documentation page.

Product and developer documentation

An agent researching an API integration could use WebMCP to search a documentation site and retrieve relevant pages. It would have a defined way to access information without navigating menus or operating the site’s search interface.

This is useful for developers working alongside browser-based agents that need accurate, up-to-date technical information. More on this below.

Customer support

An agent helping someone resolve a technical issue might need to search support resources or retrieve account information. It could also help prepare a support request.

A support website could expose tools for these actions, giving the agent a more reliable way to complete them.

E-commerce and online shopping

You could ask an agent to find running shoes under $150 that are available in your size. The agent could use WebMCP tools to search the catalog, filter products, and check availability.

The website defines the available actions, and the agent selects the tools and inputs that match your request.

Travel and bookings

Booking a trip involves choosing dates and destinations, comparing options, and entering passenger information. WebMCP could let an agent complete these steps through structured tools, reducing its reliance on interpreting individual booking forms.

Developers can design workflows that request user confirmation before sensitive actions, such as confirming a purchase.

How to add WebMCP to your website

Start by identifying which tasks agents should be able to perform, then expose those actions as structured tools.

1. Identify useful actions

Choose functionality that users might reasonably ask an agent to use. Searching content, filtering results, filling out forms, and retrieving specific information are all possible starting points.

Focus on actions with a clear purpose and predictable outcomes.

2. Expose those actions as WebMCP tools

Use HTML annotations or JavaScript, depending on the functionality you’re exposing.

Each tool should have a clear name, a description of what it does, and a structured definition of its accepted inputs. A documentation search tool, for example, might accept a search query and return matching pages.

3. Test the tools with a compatible browser agent

Check whether an agent can discover each tool, supply the correct inputs, and execute the intended action.

Review authentication and permissions as part of this work. Actions that access sensitive data or make changes also need appropriate user confirmation.

WebMCP remains experimental, and browser support is limited. Chrome provides experimental support through development features and an origin trial. Check the latest implementation guidance before getting started.

How GitBook supports WebMCP

AI assistants need access to documentation wherever people are working, including within the browser. GitBook’s native WebMCP support lets compatible browser agents search published documentation and retrieve pages in Markdown, a format AI systems can process more easily.

This works alongside GitBook’s MCP servers for published documentation, which let AI tools connect to docs outside the browser. Agents can access the content through the interface that suits their workflow.

WebMCP support is built into published GitBook sites with Connect with MCP server enabled in your site’s customization settings, so teams don’t need to build their own integration to expose these documentation tools.

Head to our announcement post to find out more about WebMCP support in GitBook, and how it works for your docs.

What does WebMCP mean for the future of websites?

Websites use menus, buttons, and forms to help people find information and complete tasks. As more users ask AI agents to handle those tasks, developers need to consider how agents will use the same functionality.

WebMCP lets websites support agents alongside their human interfaces. People can continue using the website directly, while agents have structured tools for the actions developers choose to expose.

It also complements other approaches to making websites accessible to AI systems. llms.txt helps AI systems discover relevant content, while MCP servers provide access to tools and information. WebMCP lets agents use website functionality within the browser. Together, these technologies support agents that can retrieve information and act on a user’s request.

Adoption will depend on browser support, developer implementation, and how people choose to use agents. For website owners, WebMCP offers a way to define those interactions explicitly while designing workflows that keep users in control.

Frequently asked questions about WebMCP

Is WebMCP the same as MCP?

No. MCP connects AI applications to external tools, services, and data sources. WebMCP is a proposed web standard that lets websites expose tools to agents operating within the browser. They share similar principles, serve different use cases, and can be used together.

Is WebMCP an official web standard?

Not yet. WebMCP is being developed through the W3C Web Machine Learning Community Group, with contributions from engineers at Google and Microsoft. The specification is a draft Community Group Report, not an official W3C Recommendation, and it isn’t on the W3C Standards Track. Its implementation may change as development continues.

Which browsers support WebMCP?

Chrome provides experimental WebMCP support through development features and an origin trial. WebMCP isn’t yet a universally supported browser capability. Check the official WebMCP documentation for current browser compatibility and testing guidance.

Does WebMCP help with SEO?

WebMCP doesn’t directly improve search engine rankings. It helps AI agents discover and use website functionality; it doesn’t change how search engines crawl or rank pages. Supporting agent interactions may become more useful as people turn to AI assistants to complete tasks online.

Can I use WebMCP without an MCP server?

Yes. Websites can expose WebMCP tools through browser APIs without a separate MCP server. The two technologies complement each other: an MCP server can provide access from external AI applications, while WebMCP supports agents using your website in the browser.

How do I test WebMCP readiness?

To test your website’s WebMCP readiness, check whether compatible agents can discover and use its exposed tools. In a browser with experimental WebMCP support, use developer tooling to inspect the registered tools, then test whether an agent can invoke them successfully with the expected inputs.

If your website doesn’t expose WebMCP tools yet, review its forms and JavaScript functionality to identify useful actions you could make available to agents.

→ Introducing WebMCP support: your docs’ MCP server, now in the browser

→ MCP explained: What is an MCP server and why it matters for documentation

→ What is llms.txt? Why it’s important and how to create it for your docs