TLDR: Microsoft and Google have introduced WebMCP, a new open standard that allows web developers to precisely control how AI agents interact with their websites using client-side JavaScript. This protocol enables developers to expose specific ‘tools’ or actions that AI agents can invoke, moving beyond agents inferring actions from a page’s visual layout. WebMCP is a client-side implementation of the broader Model Context Protocol (MCP), which has seen significant adoption across major AI platforms in 2025.
In a significant advancement for web development and artificial intelligence integration, Microsoft and Google have jointly unveiled WebMCP, a novel open standard designed to give developers granular control over how AI agents engage with web pages. Announced on September 30, 2025, WebMCP leverages client-side JavaScript to define explicit actions that AI agents can perform, streamlining interactions and enhancing reliability.
WebMCP, which stands for Web Model Context Protocol, is described as a set of JavaScript functions that allow web developers to ‘list the actions (called “tools”) that an AI agent can perform on a page, as JavaScript functions registered via a browser API.’ This approach contrasts with traditional methods where AI agents might attempt to understand a webpage’s structure or functionality by analyzing its Document Object Model (DOM) or visual screenshots. Instead, WebMCP provides a structured, API-driven mechanism for interaction.
Kyle Pflug, group product manager for the web platform at Microsoft Edge, highlighted the protocol’s intent in an email interview, stating that ‘MCP provided a natural starting point, but requires new platform-layer tools to cover human-in-the-loop agentic browsing scenarios.’ Patrick Brosset, a colleague of Pflug, further elaborated in a blog post, introducing WebMCP as ‘a proposal to let you, web developers, control how AI agents interact with your web pages.’
The core idea is to transform web pages into client-side Model Context Protocol (MCP) servers. MCP itself has emerged as a foundational standard in 2025 for connecting AI systems to external resources and tools. Major players like OpenAI and Microsoft have integrated MCP into their offerings, with OpenAI adopting it in March 2025 for products like ChatGPT desktop app and its Agents SDK, and Microsoft announcing general availability in Copilot Studio in May 2025. AWS also introduced MCP servers for its Lambda, ECS, EKS, and Finch services in May 2025, signaling its maturation into a production standard.
For developers, WebMCP offers the power to explicitly choose which actions to expose, how they are named and described, and their underlying functionality. This includes the flexibility to register and unregister actions dynamically to reflect changes in a page’s state. For instance, a media player web application could register actions such as ‘play,’ ‘pause,’ ‘next,’ ‘previous,’ or ‘search,’ allowing an AI agent to invoke these functions directly through the browser. Similarly, a to-do list application could expose ‘create,’ ‘edit,’ ‘complete,’ and ‘delete’ task actions.
Also Read:
- The Rise of Agentic AI: New Integrations Drive Innovation Across Industries
- Wallarm Spearheads New A2AS Standard for Enhanced Agentic AI Security
While WebMCP empowers developers with control, it does not preclude AI agents from utilizing other methods for page analysis or interaction. However, by providing a standardized and explicit interface, WebMCP aims to make AI agent interactions more efficient, reliable, and predictable, marking a significant step in the evolving landscape of AI-driven web experiences.
