The internet is currently designed for human eyes—full of buttons to click and menus to navigate. But as AI agents take over tasks like booking flights or filing support tickets, the “old way” of browsing is becoming a bottleneck. Enter WebMCP, Google’s new backbone for the “agentic web.” Recently shipped via Chrome 146 Canary, this protocol changes how AI assistants talk to websites, moving away from messy screen-scraping toward a structured, machine-to-machine conversation.
What is WebMCP?
WebMCP (Web Model Context Protocol) is a browser-based standard that allows websites to provide a “tool contract” to AI agents. Instead of an AI “looking” at a screenshot of a website and guessing where the “buy” button is, the website explicitly tells the browser: “I have a function called buyTicket that requires a date and a destination.”
-
The API: Accessible via
navigator.modelContext. -
Efficiency: Reduces computational overhead by 67% compared to visual-based agents.
-
The creators: Developed by Google (led by Alex Nahas) with collaboration from Microsoft.
Why do we need it?
Currently, AI agents are “brittle.” If a website changes its layout or moves a button, the agent breaks. Agents usually operate in two inefficient ways: 1. Visual processing: Taking constant screenshots and “reading” them (slow and very expensive in terms of data/tokens). 2. DOM scraping: Parsing raw HTML code to find elements (unreliable and messy). WebMCP solves this by creating a stable, structured interface that doesn’t care about the visual design of the site.
How it works: Two paths for developers
Developers can integrate WebMCP into their sites using two different methods:
-
Declarative API: Uses standard HTML forms with new metadata tags like
toolname. Best for simple actions (e.g., signing up for a newsletter). -
Imperative API: Uses JavaScript to handle complex, multi-step logic. Best for dynamic workflows (e.g., configuring a custom laptop or complex booking).
The “agentic web” strategy
WebMCP isn’t an isolated project; it’s part of a larger roadmap Google is building to automate the internet:
-
Universal Commerce Protocol: Standardizes how agents handle shopping and checkouts.
-
WebMCP: The layer below that handles how agents interact with any generic website.
-
Agent-to-agent protocol: Handles fully autonomous “bot-to-bot” communication (whereas WebMCP is designed for “collaborative” browsing where a human is still present).
Privacy and control
One of the biggest concerns with AI agents is them “running amok” and spending money or sharing data without permission.
WebMCP is built for scenarios where the user approves actions. The browser (Chrome) acts as a mediator, prompting the user before the agent executes sensitive operations.
Websites must explicitly choose to expose these tools; agents cannot just take over a site.
Why this matters for the future
For businesses, this is the new SEO (search engine optimization). In the near future, being “discoverable” by a human won’t be enough; you’ll need to be “usable” by an AI. Websites that adopt WebMCP will likely see more conversions because AI agents, acting on behalf of users, will find it easier and cheaper to complete transactions on those sites compared to “unstructured” competitors.
