Google Releases WebMCP for Early Preview

Google just released WebMCP for early preview, and buried in the technical documentation is a fundamental shift in who controls web interactions: not you, not the website owner—the AI agent.

On February 10th, Google announced WebMCP (Model Context Protocol for the web), proposing two new APIs that allow browser agents to "take action on behalf of the user." The pitch sounds reasonable: structured interactions for booking flights, filing support tickets, and navigating e-commerce. The reality is that Google is standardizing how AI agents bypass your actual website to perform actions you never explicitly designed or authorized.

This isn't innovation. This is annexation.

What WebMCP Actually Does

WebMCP introduces two APIs that fundamentally change how agents interact with websites:

Declarative API: Performs "standard actions" defined directly in HTML forms. Translation: AI agents can fill out and submit forms without rendering your actual website interface, using structured definitions you're expected to provide.

Imperative API: Executes "complex, more dynamic interactions that require JavaScript execution." Translation: AI agents can trigger JavaScript functions and perform multi-step workflows by directly manipulating your site's functionality.

Google frames this as making websites "agent-ready" and enabling "more reliable and performant agent workflows compared to raw DOM actuation." What they're not saying: this replaces website owners' control over user experience with Google's control over how agents access your functionality.

The use cases Google highlights—customer support ticket creation, ecommerce product configuration, travel booking searches—all involve actions that currently require users to interact with carefully designed interfaces. WebMCP proposes that agents should bypass those interfaces entirely.

The Control You're Expected to Surrender

Here's what WebMCP actually requires from website owners:

You define structured tools that tell agents "how and where to interact with your site." You provide the schemas, the endpoints, the structured data that makes your site machine-readable. You essentially build a parallel API layer specifically for AI agents—maintained separately from your actual user interface—or risk agents performing "raw DOM actuation" (automated clicking and form-filling) that you can't predict or control.

Google presents this as eliminating ambiguity and enabling "faster, more robust agent workflows." But ambiguity serves a purpose. When users interact with your website, you control pacing, presentation, upsells, cross-sells, brand experience. You can A/B test flows, implement progressive disclosure, guide users through complex decisions.

WebMCP removes all of that. Agents get direct access to functionality without experiencing your interface. They extract value—booking confirmations, support ticket numbers, product purchases—without engaging with anything you built around those transactions.

Who Benefits From "Agent-Ready" Websites?

Not website owners. You're expected to invest engineering resources to implement WebMCP, maintain structured definitions as your site evolves, and debug when agents misuse your APIs. In exchange, you get traffic that bypasses your interface, never sees your brand, and attributes the entire interaction to the AI agent rather than your business.

Not users. The promise is "helping users create detailed customer support tickets" by "enabling agents to fill in all of the necessary technical details automatically." But customer support tickets require context, nuance, description of attempted solutions. Automated form-filling doesn't improve support—it floods support queues with agent-generated tickets lacking the human context that makes them useful.

Google benefits. Every website that implements WebMCP becomes more valuable to Google's agents (and by extension, Google's ecosystem). Search stops being about finding websites and becomes about agents extracting structured actions from WebMCP-enabled sites. The web becomes infrastructure for Google's agent platform.

The Ecommerce Example Nobody's Examining

Google's travel and ecommerce use cases deserve scrutiny. "Users could more easily get the exact flights they want, by allowing the agent to search, filter results, and handle bookings using structured data." This isn't about ease—it's about disintermediation.

Currently, when users book flights on Expedia, Kayak, or airline websites, those companies control the search experience. They can promote certain routes, highlight partnerships, offer upgrades, cross-sell hotels and rental cars. That's their business model. That's how they monetize traffic.

WebMCP proposes that agents bypass all of it. The agent searches, filters, and books using "structured data to ensure accurate results every time." The airline or booking platform becomes a commodity backend API. The agent—and by extension, the agent platform—captures the customer relationship.

For ecommerce: "Users can better shop your products when agents can easily find what they're looking for, configure particular shopping options, and navigate checkout flows with precision." Precision means bypassing your product discovery, your recommendations, your curated shopping experience. The agent finds the product, configures options, completes checkout—and the customer never sees your brand.

According to Statista, global ecommerce revenue exceeded $6.3 trillion in 2024. That entire industry is built on controlling customer experience, optimizing conversion funnels, and building brand loyalty through interface design. WebMCP proposes to eliminate all of it.

The "Early Preview" That's Already Too Late

WebMCP is "available for prototyping to early preview program participants." You can "sign up for the early preview program to gain access to the documentation and demos, stay up-to-date with the latest changes, and discover new APIs."

Notice what's missing: consultation. Google isn't asking website owners whether they want this. They're not proposing a standard through W3C or any multi-stakeholder process. They're building it, deploying it, and inviting you to "gain access" to documentation for a system that fundamentally changes how agents interact with your infrastructure.

The Model Context Protocol (MCP) was originally developed by Anthropic for connecting AI systems to data sources. It's designed to give AI agents structured access to tools and information. WebMCP extends that concept to the entire web—without asking whether the web wants to be restructured as agent-accessible tooling.

What This Actually Costs

Implementing WebMCP isn't free. You need to:

• Define structured tool schemas for every interactive element on your site
• Maintain those definitions as your site evolves
• Test agent interactions separately from user interactions
• Debug when agents misuse APIs or perform unintended actions
• Monitor agent traffic separately from human traffic
• Build rate limiting and fraud detection for agent-driven actions

And you get what in return? The privilege of having Google's agents extract value from your site without users engaging with your actual product.

For small and medium businesses, this is particularly brutal. You don't have engineering teams to maintain parallel API layers for agents. You can't compete with companies that implement WebMCP early. And when Google's agents start prioritizing WebMCP-enabled sites in their responses, you'll be forced to implement it just to maintain visibility.

This is the same playbook Google used with AMP (Accelerated Mobile Pages). Implement our standard or lose search visibility. Except WebMCP is worse—AMP at least rendered your content. WebMCP extracts functionality without showing your interface at all.

The Compatibility Problem Nobody's Addressing

WebMCP proposes standardized APIs, but every website is different. Your customer support form might require context that doesn't fit the structured schema. Your ecommerce flow might have custom steps for compliance, verification, or brand experience. Your booking system might need human judgment calls that agents can't make.

Google's answer: define those complexities in structured formats that agents can understand. But structured formats don't capture human judgment. They don't handle edge cases. They don't adapt to situations that the schema designer didn't anticipate.

When agents inevitably misuse these APIs—booking wrong flights, creating malformed support tickets, purchasing incorrect products—who's responsible? The website owner whose API "failed" to prevent misuse? The agent that misinterpreted the schema? Google, whose platform enabled the entire interaction?

What We Should Actually Do

Website owners need to understand that "agent-ready" means "giving up control." Before implementing WebMCP:

• Calculate the actual cost: engineering time, maintenance burden, opportunity cost of building agent interfaces instead of user interfaces
• Assess whether direct agent access actually benefits your business model or undermines it
• Determine if you can realistically prevent agent misuse through schema design alone
• Consider whether losing the customer relationship to the agent platform is worth the traffic

For businesses whose value proposition includes interface design, curation, brand experience, or guided discovery—which is most businesses—WebMCP is a trap. You're being asked to dismantle your competitive advantages so agents can extract your functionality more efficiently.

The web evolved as a network of human-readable interfaces controlled by their creators. WebMCP proposes transforming it into a machine-readable infrastructure controlled by agent platforms. That's not progress. That's consolidation of power under the guise of standardization.

Don't let platform standards undermine your business model. Winsome Marketing's growth experts help you evaluate emerging technologies like WebMCP through the lens of actual business impact, not vendor promises. Let's talk about building web strategies that preserve your competitive advantages.