# The Rise of WebMCP

Source: https://www.youtube.com/watch?v=35oWt7u2b-g
Recap page: https://rapidrecap.app/video/35oWt7u2b-g
Generated: 2026-02-12T14:04:13.573+00:00

---
## Quick Overview

The presentation introduces WebMCP (Web Machine Context & Procedure), a new static JSON document standard designed to enable AI agents to interact with websites more reliably and efficiently by exposing structured tools and context directly via the browser, rather than relying on costly, brittle methods like DOM manipulation or screen scraping.

**Key Points:**
- WebMCP (Web Machine Context & Procedure) is proposed as a new standard to allow websites to expose structured tools directly to AI agents.
- The system relies on three pillars for Agent-User-Site Interaction: Context (relevant user state), Capabilities (ability to take actions on user's behalf), and Coordination (interplay between users and agents on the same UI).
- WebMCP defines two new APIs for structured interactions: the Declarative API for standard actions defined in HTML forms, and the Imperative API for complex, dynamic interactions requiring JavaScript execution.
- This approach aims to reduce token costs and improve reliability compared to existing methods like raw HTML parsing or relying on thousands of screenshots per task.
- Use cases for WebMCP include Customer Support (auto-filling tickets), E-commerce (precise shopping flows), and Travel (accurate flight searches).
- Google and Microsoft collaborated on this specification, which was presented at the WebAI Summit, aiming for wider adoption in the coming months.

![Screenshot at 00:27: The slide illustrating the three pillars of Agent-User-Site Interaction: Context, Capabilities, and Coordination, which form the foundation of the WebMCP standard.](https://ss.rapidrecap.app/screens/35oWt7u2b-g/00-00-27.jpg)

**Context:** This video presents a technical proposal, WebMCP (Web Machine Context & Procedure), developed by contributors from Google Chrome, aimed at standardizing how AI agents interact with websites. The speaker contrasts the proposed method with older, less reliable techniques like screen scraping or heavy DOM manipulation, framing WebMCP as a solution for building more robust, efficient, and cost-effective AI agents for web automation.

## Detailed Analysis

The presentation argues that current methods for AI agents interacting with websites—such as screen scraping, parsing raw HTML, or relying on extensive visual data (screenshots)—are expensive in terms of tokens, slow, and brittle. The proposed solution is WebMCP (Web Machine Context & Procedure), a static JSON document that allows websites to explicitly declare structured tools and context for AI agents. The core concept is Agent-User-Site Interaction, built on three pillars: Context (the user's current state), Capabilities (the ability for the agent to take actions on the user's behalf, moving beyond simple help to task completion), and Coordination (managing the interplay between user and agent on the same UI). WebMCP introduces two new APIs: the Declarative API, which handles standard actions defined directly in HTML forms (like submitting a form), and the Imperative API, which handles complex, dynamic interactions requiring JavaScript execution. The benefit is providing agents with a structured, unambiguous interface, potentially reducing reliance on costly visual processing. Use cases highlighted include customer support (auto-filling tickets), e-commerce (precise checkout flows), and travel booking. The speaker notes that Google and Microsoft have been working on this specification, suggesting it is gaining traction for future browser support.

### Agent-User-Site Interaction Pillars

- Context: The contextual information relevant to user's current state
- Capabilities: The ability to take actions on the user's behalf, moving from helping to doing tasks for them
- Coordination: The interplay of users and Agents interacting with the same UI

### WebMCP APIs

- Declarative API: Perform standard actions that can be defined directly in HTML forms
- Imperative API: Perform complex, more dynamic interactions that require JavaScript execution

### Comparison to Existing Methods

- Eliminates the need for scraping/parsing raw HTML or using thousands of screenshots per task, reducing token costs and increasing reliability.

### Agent Tool Definition Example

- Shows imperative API registration for an 'add-to-cart' tool with defined input schema (object with productID and quantity) and a declarative API example for an 'add-todo' form.

### Use Cases

- Customer support (auto-filling detailed support tickets)
- Ecommerce (better shopping via precise configuration and checkout flows)
- Travel (accurately searching and filtering flights)

![Screenshot at 00:03: The initial view showing the ChatGPT-like interface where an agent is being used to navigate to vrbo.com.](https://ss.rapidrecap.app/screens/35oWt7u2b-g/00-00-03.jpg)
![Screenshot at 00:18: A screenshot showing the WebUI interface running an agent, highlighting the inputs for agent description and running the agent.](https://ss.rapidrecap.app/screens/35oWt7u2b-g/00-00-18.jpg)
![Screenshot at 01:33: A code snippet demonstrating the structure of registered tools for both Imperative API \(e.g., 'add-to-cart'\) and Declarative API \(e.g., 'add-todo'\).](https://ss.rapidrecap.app/screens/35oWt7u2b-g/00-01-33.jpg)
![Screenshot at 02:28: A timeline graphic illustrating the progression of web interaction technologies, ending with Chrome 146 support.](https://ss.rapidrecap.app/screens/35oWt7u2b-g/00-02-28.jpg)
![Screenshot at 05:07: A slide detailing the three pillars of Agent-User-Site Interaction: Context, Capabilities, and Coordination.](https://ss.rapidrecap.app/screens/35oWt7u2b-g/00-05-07.jpg)