WebMCP: Stop scraping the Web, start calling it
summary
Demo shifts from brittle DOM scraping to explicit page-exposed tools: a pizza-order scenario goes from a slow, click-and-screenshot agent to a direct tool call that completes checkout and consumes half as many tokens. WebMCP offers a declarative HTML path and a JavaScript API (register via document.mmodel.context.registerTool) so site code handles state and UI logic, removing UI flakiness. Spec advances at W3C, works in Edge behind a flag and has early support from OpenAI, Cloudflare, Puppeteer and Playwright.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 0:30Web’s dual role
- 2:17Limitations of current agentic browsing
- 6:55Websites as participants
- 8:19Introducing WebMCP
- 9:46WebMCP as API
- 10:51Adopting WebMCP
- 12:12Benefits over scraping
- 14:45Why not MCP
- 16:32WebMCP for non-experts
- 17:34WebMCP status
- 18:38Ecosystem adoption
- 19:19Call to action
related talks
23mBeyond Scraping: The Future of Web Data Access with WebMCP
22mThe agent-ready web: Simplify user actions with WebMCP
11mDon't let AI agents push your buttons - use webMCP instead!
23mIntroducing WebMCP: Agents in the Browser
30mPractical steps to make your website agent-ready
45mModel Context Protocol (MCP): The future of user interactions?
22mMCP UI: Extending the frontier
53mAI Didn’t Kill the Web, It Moved in!