Google Releases WebMCP: Turning Websites into AI Agent Interfaces

AI agents are becoming the internet's new users—is your website ready? Google's newly released experimental feature WebMCP attempts to answer this question.

From Screenshots and Clicks to Protocol Conversations: The Evolution of AI Browsing

Over the past few years, major model providers have been trying to get AI agents to browse the web like humans. According to analysis in the YouTube video, the current mainstream approach has agents open a page, take a screenshot, determine cursor position, click, wait, screenshot again to evaluate—a process that is both slow and expensive. While this "visual simulation" path is universal, it is extremely inefficient.

Anthropic's MCP (Model Context Protocol), launched in 2025, changed the game. It allows developers to establish API-like interfaces between applications and AI agents, freeing agents from having to "look at the screen" to operate.

What is MCP (Model Context Protocol)?

MCP is an open protocol proposed by Anthropic in late 2024 and gradually promoted throughout 2025, short for Model Context Protocol. Its core idea is to define a standardized communication specification between AI models and external tools and data sources—similar to what USB-C is for hardware peripherals: as long as both sides follow the protocol, it's plug and play.

Before this, tool-calling solutions were typically function calling formats defined separately by each model provider, incompatible with one another. MCP abstracts this layer into a unified client-server structure: applications expose capabilities as MCP Servers, and AI agents act as Clients to discover and invoke them on demand. This not only reduces integration costs but also allows agents to dynamically retrieve tool lists at runtime without hardcoding interfaces in prompts. WebMCP is precisely about pushing MCP's Server side down into the browser and the webpage itself, making every website a natural capability node that agents can invoke.

What WebMCP Does: Turning Websites into MCP Servers

Google's WebMCP extends this idea to entire websites. According to the video, website owners only need to add a small amount of code to interactive elements, and the site instantly becomes an MCP server, dramatically lowering the difficulty of AI agent-website interaction. The feature is currently experimental only within the Chrome browser, but the video author believes its logic is clear enough that widespread adoption is only a matter of time.

Key insight: WebMCP essentially transforms "humans finding features" into "agents directly invoking features." For developers, this may be the third interaction paradigm shift websites need to face, after mobile adaptation and SEO.

What does "adding a small amount of code to interactive elements" mean?

Traditional webpage interactive elements (buttons, forms, dropdowns) only have meaning for rendering engines and human accessibility tools (such as screen readers). WebMCP's approach is to attach structured semantic descriptions to these elements—for example, "this button submits an order," "this input field accepts a zip code," "this endpoint returns inventory quantity"—and declare their parameters and return formats using schemas defined by MCP.

This way, AI agents don't need to understand pixel layouts or DOM hierarchies; they simply read this declaration to know "what can I do on this page and what parameters should I pass." This follows the same lineage as using structured data (Schema.org, JSON-LD) in SEO to help search engines understand page content, with the difference being that the target reader shifts from crawlers to agents that can actively perform operations. For developers, this means webpages must not only consider being "read and understood" but also being "invoked."

Why Your Website Needs to Cater to AI Agents

The video raises a question worth pondering for all website owners: why make it easier for AI to operate my website? The answer is straightforward—humans are using websites less frequently, and letting AI handle things is increasingly convenient. If AI cannot interact smoothly with your website, it will turn to competitors' sites.

Data supports this trend: information cited in the video shows that AI traffic has recently surpassed human traffic and is projected to reach about 7.5 times human traffic by this time next year.

This means a website's "AI-friendliness" may soon matter more than its "human-friendliness." The video author suggests you don't need to integrate WebMCP immediately, but should closely monitor its future development.

The measurement of "AI traffic" statistics requires careful interpretation

The claim cited in the article that "AI traffic has surpassed human traffic and will reach 7.5 times next year" is a highly viral but nuanced statement. "AI traffic" here typically refers to visits generated by automated agents, crawlers, and model training and inference requests—not equivalent to the volume of real user behavior where "AI agents complete transactions on behalf of humans."

The difference between the two is enormous: a crawler fetching a page versus a user completing a purchase have completely different business value. A more accurate understanding is therefore that network requests initiated by machines have long overwhelmed human clicks in magnitude, and the new type of requests brought by AI agents (which have goals, execute operations, and may represent genuine user intent) are becoming the fastest-growing segment. What website owners really need to assess is: among these machine visits, how many can convert into actual business value—rather than simply chasing the number of "visits."

Opportunities and Unresolved Questions

WebMCP is still in an early experimental stage, with implementation details, security considerations, and cross-browser support all unclear. The video author also mentions plans to follow up with more information.

From a broader perspective, WebMCP represents a trend: AI agents are no longer "second-class citizens" of the internet but are becoming its primary users. Website owners need to start thinking about how to serve both human users and AI agents well—this may be one of the most important topics in web development over the coming years.

One-line Summary

Google WebMCP turns websites into MCP servers, with AI agents poised to become the internet's primary users.

Paragraph Summary

Google's experimental WebMCP feature aims to let websites become MCP servers by adding a small amount of code, enabling AI agents to directly invoke site functions without relying on inefficient visual simulation like screenshots, cursor movement, and clicks. This approach builds on Anthropic's MCP protocol, essentially transforming "humans finding features" into "agents directly invoking features," and is seen as the third interaction paradigm shift websites face after mobile adaptation and SEO. The article notes that AI traffic has already surpassed human traffic and is expected to keep climbing—if a website can't be operated smoothly by AI, users may flow to competitors. Although WebMCP is currently experimental only in Chrome and security and cross-browser support remain unclear, website owners should start paying attention to AI-friendliness as a new challenge.