The short answer to what is webmcp: it is a JavaScript and HTML proposal that gives web pages a standard way to tell an AI agent which actions it can take on the page, what each action needs as input and what it returns. The name stands for Web Model Context Protocol. It borrows ideas from the Model Context Protocol (MCP), which Anthropic published in 2024 to connect AI applications to tools and data, but it runs inside the browser tab the user already has open.
Why AI agents need something like WebMCP
AI agents that browse on a user's behalf mostly work by reading the screen or the page code and simulating clicks and typing. That route is slow and fragile: a redesigned button, a cookie banner or a multi-step form can break the task halfway. WebMCP proposes a direct route. The site publishes its actions as tools, each with a name, a description in plain language and a schema for its inputs, and the agent calls them like functions. A flight search becomes a tool that takes origin, destination and date; a quote request becomes a tool that takes service, location and contact details.
Asking what is webmcp is really asking how agents will complete tasks on websites: by imitating a person, or by using an interface the site designed for them.
How WebMCP works: declarative and imperative tools
WebMCP works through a new browser interface called navigator.modelContext. The proposal is being incubated in the W3C Web Machine Learning Community Group, with engineers from Google and Microsoft among its editors, and it describes two ways to register tools:
| Approach | How it works | Suited to |
|---|---|---|
| Declarative | Existing HTML forms receive extra attributes with a tool name and description, and the browser turns the form into a tool | Contact, quote, search and booking forms that already exist |
| Imperative | JavaScript registers each tool with a name, a description, an input schema in JSON Schema and a function that runs when called | Dynamic actions such as filtering a catalog, adding to a cart or configuring a product |
In both cases the tool runs in the page with the user's existing session, so the agent acts with the same permissions the user already has, and no separate server or login is needed. The explainer also describes ways for a tool to ask the person for confirmation before a sensitive step, which keeps a human in the loop.
What is webmcp's status in 2026?
WebMCP is not a finished standard. Chrome made an early preview available to developers behind a flag in early 2026, and the API names and attributes can still change before any stable release. There is no stable support across browsers yet. No search engine has said it uses WebMCP for ranking, and Google Search does not list it among the signals in its documentation, so any claim that WebMCP improves rankings should be treated with suspicion.
WebMCP compared with MCP, llms.txt and schema markup
Four names often get mixed up in conversations about AI and websites. Each solves a different problem:
| Technology | What it does | Where it lives | Status |
|---|---|---|---|
| MCP | Connects AI applications to tools and data through servers | A server or local process | Open protocol published by Anthropic in 2024 |
| WebMCP | Exposes page actions to agents in the browser | Inside the web page | Proposal in early preview |
| llms.txt | Summarizes a site and lists key pages in Markdown for language models | A text file at the site root | Proposal by Jeremy Howard, 2024; Google does not use it |
| Schema.org markup | Describes entities and content in a structured way | JSON-LD in the page | Established vocabulary used by search engines |
The distinction matters for planning: schema markup and llms.txt help systems understand what a page is, while WebMCP helps an agent do something on it.
What should a business do about WebMCP now?
The practical side of what is webmcp for most companies is preparation, not implementation. These steps pay off for users and accessibility today and make a future rollout simpler:
- Build forms with proper labels, clear field names and readable error messages, the same structure declarative tools would rely on.
- List the actions that matter on your site, such as requesting a quote, booking a call, checking availability or finding a product. Those are the candidate tools.
- Keep core content readable in the HTML, since agents and AI crawlers still need to understand the page before acting on it.
- Let developers test the Chrome preview on a staging copy, never on production flows that handle payments or personal data.
- Keep investing in SEO, GEO and AEO, because an agent only reaches your site after a search engine or an assistant points to it.
Being found comes before being usable. Before an agent fills in your quote form, an AI system has to recommend your business, which is the job of generative engine optimization services. Our guides explain how generative engine optimization works and what answer engine optimization adds. Through our AI SEO agency service, our team in Brazil works remotely with businesses in the United States and Ireland, applying a method built over more than 700 clients served in Brazil since 2021.
Want a clear read on how ready your site is for AI search and for browser agents? Talk to Busca Orgânica and we will review your pages, forms and structured data with you in a call.