Descriptions:
Greg Isenberg sits down with developer Vinnie to unpack WebMCP, a new experimental protocol from Google and Microsoft that lets websites expose structured, agent-readable actions — think search, filter, and buy endpoints that an AI agent can call directly instead of scraping and clicking through the DOM like a human.
The episode opens with a live demo on a real espresso machine e-commerce store, where an agent using WebMCP narrows down compatible machines based on counter width and daily brewing volume — a task that would have required clunky computer-use screenshot scraping before. Vinnie draws a clear three-way distinction: legacy browser agents that parse raw DOM, in-app agents where the vendor controls the model and pays for tokens, and WebMCP as the pragmatic middle ground where the user brings their own agent (Claude Code, ChatGPT, Grok) and the site simply publishes its capabilities.
Isenberg closes with two cash-flow startup ideas built on WebMCP, framing the protocol as a first-mover opportunity as the broader web begins to redesign itself for AI visitors. The conversation is accessible for builders curious about what “agent-ready” infrastructure actually looks like in practice today.
📺 Source: Greg Isenberg · Published August 26, 2026
🏷️ Format: Podcast
