
The podcast introduces WebMCP, a new specification that allows websites to surface tools and functionalities for AI interaction directly through the website itself, contrasting it with traditional MCP servers and UI approaches. Wes demonstrates how WebMCP enables AI to interact with a website, using a grocery list app as an example, where AI can add, remove, and reorder items via natural language commands. The implementation involves publishing tools either declaratively via HTML forms or imperatively with JavaScript, streamlining AI interaction by providing structured input/output schemas. Wes highlights WebMCP's speed and token efficiency compared to AI browser interactions that rely on parsing HTML or screenshots. They discuss the potential for frameworks to adopt WebMCP and consider the possibility of cross-application interactions.
Sign in to continue reading, translating and more.
Continue