What Is WebMCP and Why Does It Matter?

What is WebMCP?

In a nutshell, WebMCP, or Web Model Context Protocol, is an emerging technical protocol that makes websites “agent-ready.” Agents interact with websites differently from human visitors, and it’s inefficient for them to get information in the same way a customer might. For instance, a person sees a button and knows to click; an agent must parse through noisy code to know where the button is, what it does, and take action on it. Because agents are taking independent actions on behalf of people, they need a more efficient, structured way to understand what they can do on a website.

WebMCP is being developed in collaboration with engineers from Microsoft, and it represents a massive shift in how AI agents interact with websites. It offers a robust alternative to what agents currently use, which is a brittle, inefficient, and expensive approach that relies on visual scraping. Rather than burn through resources to parse screenshots or raw HTML to determine where a button is and what it does, WebMCP allows websites to provide a clear structure for their functions — like searching and filtering travel results, navigating the checkout process, or creating customer support tickets — via a new browser API, called navigator.modelContext.

On February 10th, 2026 Google announced an early preview of WebMCP within Chrome 146 Canary. If you’re interested in exploring it, you can apply for the preview to WebMCP here.

A simple definition of WebMCP

If HTTP is the language browsers use to fetch pages, WebMCP is the language AI agents use to call tools. It operates on top of the web, not as a replacement for HTTP. The goal is to provide a structured way for websites to explicitly say to an AI: “Here is a list of things I can do, here are the inputs required, and this is the output when you ask.” This eliminates the guesswork and provides a frictionless way for an agent to book a hotel room, query inventory, or reserve a table at a specific restaurant.

Implementation overview

Under the hood, WebMCP follows a straightforward three-step process that maps well to existing AI workflows:

1. Discovery: 

The agent first asks “What tools can be utilized on this site?” The website then responds with a manifest of tool names, natural-language descriptions, structured input, and expected output schemas. 

2. Capabilities: 

Once the manifest has been received, the AI then reviews to see capabilities like get_inventory, book_appointment, create_invoice, or apply_coupon. Each requires clearly typed parameters and return values.

3. Execution: 

The agent, now knowing what is available to it, will then select the appropriate tool, fill in the arguments in a structured format (e.g. JSON), then execute it. The site runs its existing logic and returns a structured result that the model can reason about and chain into a larger workflow.

The most important consideration with WebMCP is that you’re not rebuilding your product or website. You’re defining the most important elements with a standardized, agent-friendly methodology. For technical SEOs, this paradigm will sound very familiar. 

Declarative vs Imperative APIs

WebMCP offers two standard methodologies for implementation, allowing teams to choose what is right for their website. 

Declarative (markup-driven): 

The Declarative API allows for standard actions that can be defined directly in HTML forms. For example, this could be used to inform an AI agent how to book a restaurant reservation. A developer would annotate a reservation form, specifying the required fields and constraints ("this creates a reservation"). The AI agent can then submit the form safely without any additional custom scripting.  

Imperative (code-driven): 

The Imperative API is designed for more complex, dynamic interactions that require JavaScript execution. This allows the AI agent to cleanly navigate multi-step processes, validations, and potential side effects. By exposing explicit tool functions in code, developers can encapsulate existing rules and workflows while presenting a highly stable interface to the agent.

Together these APIs act as a bridge, allowing your website to become “agent-ready.”

WebMCP vs structured data (Schema.org)

It may seem odd to mention Schema.org, but the last time we saw Google and Microsoft team up to release an open standard, it fundamentally changed technical SEO. The two concepts are intimately related, as both elements are designed to provide clarity to machines. 

Structured data can be thought of as the way that we explain what things are to an AI. It says “This is a product, this is its price, these are the reviews, these are the frequently asked questions, etc.” It describes the nouns on the page. 

WebMCP, on the other hand, focuses on the actions that an AI can take. It tells agents what they can do on the site — whether it’s a purchase, subscribing to a newsletter, or booking a hotel room. Think of WebMCP as the way to describe the verbs on your site. 

These elements together allow agents not just to understand where to go and what they’re looking at, but how to effectively take a useful action. WebMCP provides the missing context for an AI: not just what is on a site, but how to use it. 

The agentic funnel: Where WebMCP meets UCP

To fully grasp the future of the agentic web, we have to look at WebMCP alongside another major Google initiative: the Universal Commerce Protocol (UCP). If WebMCP makes your website "talk" to an AI, UCP is what allows the AI to take meaningful actions to authorize payments and manage transactions safely.

They represent Step 1 and Step 2 of the new agentic funnel:

• Step 1: Context & interaction (WebMCP)
  The AI uses WebMCP to understand the site, filter inventory, apply logic, and set up the conversion (e.g., adding items to a cart or initiating a booking).
• Step 2: Transaction & fulfillment (UCP)
  Once the consumer is ready to buy, UCP takes over. It handles the standardized checkout experience, payment handler routing (like Google Pay), and post-purchase order lifecycles natively within the AI surface.

Together, WebMCP handles the conversation, UCP handles the conversion.

The bottom line:

This is a massive leap forward for the agentic future, making AI assistants infinitely more reliable and capable. If Schema.org represented a contextual shift for websites, WebMCP represents the behavioral evolution. For SEO and digital marketing teams, consider the following next steps:

1. Identify high-value actions: Audit your site for core "verbs" (e.g., book, buy, subscribe, request quote).
2. Align with development: Discuss whether Declarative or Imperative integrations make the most sense for your tech stack.
3. Monitor the rollout: Track the WebMCP spec, Chrome’s Early Preview Program (EPP) implementation status, and UCP developments.
4. Rethink analytics: Begin developing a measurement plan for "agent-completed" attribution.

To quote Dan Petrovic of DEJAN: “We’re watching the early days of a new layer of the web stack. If you’re in technical SEO, start paying attention now.”