Vince Canger demonstrates WebMCP as a browser-level way for websites to expose selected actions directly to AI agents. Instead of interpreting an entire page or repeatedly taking screenshots, an agent receives a concise tool list for tasks such as searching, comparing products, adding items to a cart or applying a coupon.
The live espresso-store example shows why the approach can be useful for complicated purchases. An agent compares machines against counter width and drink requirements, recommends compatible accessories and updates the visible shopping interface. Because the tools share the existing browser session, their availability can change when a user logs in or out without requiring a separate API credential flow.
Greg Isenberg and Canger position WebMCP between headless APIs, brittle computer use and product-specific in-app assistants. They see early applications in compatibility-driven commerce, administration tools, internal systems and restricted read-only workflows. The technology remains experimental, so its business value still depends on adoption, careful tool design and reliable evaluation of the actions agents can complete.
Watch on YouTube