WebMCP: teach Agents to browse your site

Key Concepts

• AI Agents: Autonomous software entities that utilize reasoning and memory to execute complex tasks.
• Web MCP (Model Context Protocol): A proposed web standard designed to expose structured data and site capabilities directly to AI agents.
• Declarative API: A method where the browser automatically translates existing HTML forms into actionable tools for AI.
• Imperative AI: A method allowing agents direct access to a site’s JavaScript to perform complex, custom operations.
• Token/Time Efficiency: The optimization of computational resources by reducing the need for agents to "scrape" and interpret human-centric UI.

---

The Evolution of AI Web Interaction

Currently, AI agents interact with the web by "scraping" screens—simulating human behaviors like clicking, scrolling, and waiting for animations to load. While functional, this approach is inefficient, consuming significant time and computational tokens. The core argument presented is that the web, which was built exclusively for human consumption, requires a new standard to facilitate machine-to-machine communication.

Introducing Web MCP

Web MCP is proposed as a standard to bridge the gap between AI agents and websites. Instead of forcing an agent to "guess" how a site works through visual observation, Web MCP allows developers to provide a structured map of the site’s capabilities.

Benefits of Implementation:

• Increased Speed: By providing a predefined list of available tools, agents no longer need to spend time discovering site functionality.
• Resource Optimization: Reduces the overhead associated with screen scraping, leading to lower token usage and faster task completion.

Implementation Methodologies

The proposal outlines two primary ways for developers to integrate Web MCP into their sites:

1. Declarative API: This approach leverages existing HTML structures. The browser interprets standard HTML forms and automatically converts them into tools that the AI agent can utilize. This is the lower-friction path for developers.
2. Imperative AI: This provides a more robust, granular level of control. Developers can grant the agent direct access to the site’s JavaScript, enabling the execution of complex, custom-defined tasks that go beyond simple form submissions.

Strategic Perspective

The transition from "human-centric" web browsing to "agent-centric" interaction is framed as a necessary evolution. By treating a website’s capabilities as a set of tools rather than a visual interface, developers can create more reliable and efficient integrations for AI assistants. The speaker emphasizes that this is an early-stage initiative, inviting developers to join the "wemp" (Web MCP) early preview program to influence the development of these APIs.

Conclusion

The shift toward Web MCP represents a move toward a more programmatic web. By moving away from visual scraping and toward structured, tool-based interaction, developers can significantly improve the performance and reliability of AI agents. The primary takeaway is that the future of web development will increasingly involve designing sites that are as readable and actionable for AI agents as they are for human users.