WebMCP is an open web standard that enables websites to expose structured, callable tools directly to browser-based AI agents. Announced by Google at Google I/O 2026 on May 21, the protocol is now available in Chrome 149 for real traffic. The specification was co-developed by Google and Microsoft and published as a draft by the W3C Web Machine Learning Community Group in February 2026. This protocol replaces error-prone methods like vision-based actuation and DOM scraping, which previously required agents to simulate clicks by guessing coordinates based on screenshots or page structures. According to 2026 implementation guides, switching to structured approaches reduces task errors by 67% and improves completion rates by 45%.

The protocol functions through the document.modelContext interface in the browser, offering two primary APIs: the Declarative API for annotating HTML forms and the Imperative API for registering dynamic, JavaScript-driven tools. By inheriting the logged-in user’s session cookies, WebMCP avoids complex OAuth requirements, ensuring the agent operates within the same permission boundaries as the user. This security model prevents privilege escalation, as agents cannot perform actions beyond what a human user is authorized to do.

A primary example provided by Google involves travel booking, where websites register structured functions like search_flights. Instead of fragile pixel-chasing or adapting to CSS changes, agents call these functions directly with precise inputs such as destination and cabin class. Developers can begin implementation by enabling chrome://flags/#enable-webmcp-testing, installing the Model Context Tool Inspector extension, and utilizing the @mcp-b/global polyfill for cross-browser support, with Microsoft Edge 147 already providing native compatibility.

WebMCP is an open web standard that enables websites to expose structured, callable tools directly to browser-based AI agents. Announced by Google at Google I/O 2026 on May 21, the protocol is now available in Chrome 149 for real traffic. The specification was co-developed by Google and Microsoft and published as a draft by the W3C Web Machine Learning Community Group in February 2026. This protocol replaces error-prone methods like vision-based actuation and DOM scraping, which previously required agents to simulate clicks by guessing coordinates based on screenshots or page structures. According to 2026 implementation guides, switching to structured approaches reduces task errors by 67% and improves completion rates by 45%.

The protocol functions through the document.modelContext interface in the browser, offering two primary APIs: the Declarative API for annotating HTML forms and the Imperative API for registering dynamic, JavaScript-driven tools. By inheriting the logged-in user’s session cookies, WebMCP avoids complex OAuth requirements, ensuring the agent operates within the same permission boundaries as the user. This security model prevents privilege escalation, as agents cannot perform actions beyond what a human user is authorized to do.

A primary example provided by Google involves travel booking, where websites register structured functions like search_flights. Instead of fragile pixel-chasing or adapting to CSS changes, agents call these functions directly with precise inputs such as destination and cabin class. Developers can begin implementation by enabling chrome://flags/#enable-webmcp-testing, installing the Model Context Tool Inspector extension, and utilizing the @mcp-b/global polyfill for cross-browser support, with Microsoft Edge 147 already providing native compatibility.