Introducing WebMCP: Agents in the Browser
summary
Claim: agents can render full interactive web apps not just chat by bundling HTML/CSS/JS as an MCP app and exposing site actions via WebMCP. Demo shows a comic reader inside Claude that preserves navigation and transcript mode; tools include getPage and getTranscript, transports compare STDIO vs HTTP, and browser agents register capabilities via navigator.modelContext.registerTools. Built artifact runs in an iframe, works for navigation and speech, and surfaces practical limits: base64 bundling, CSP, no localStorage, and server-mediated network calls.
AI-written summary · about
chapters (14)
- 0:00Start
- 0:07Introduction and Speaker Background
- 3:28Motivations for Future-Proofing the Web Comic Site
- 5:09Roadmap: MCP, Apps, and WebMCP
- 6:05Understanding Transports: STDIO vs. HTTP
- 8:57Defining MCP Tools
- 10:52The Role of MCP Resources
- 11:37The Case Against 'Starfish' (Chat-Only) UI
- 13:36MCP Apps: Building Interactive Agent Interfaces
- 16:34Best Practices and Gotchas for MCP Apps
- 17:38Introducing WebMCP: Agents in the Browser
- 19:07Imperative vs. Declarative Models
- 20:13WebMCP Demo with the Debugging Extension
- 21:47Leveraging Native Browser APIs (Speech, Animation, etc.)
related talks
49mThe Headless Web
11mDon't let AI agents push your buttons - use webMCP instead!
23mBeyond Scraping: The Future of Web Data Access with WebMCP
38mEvolving AI chat with MCP Apps
29mMCP Apps: Primitives, discovery, and the Future of Software
20mWebMCP: Stop scraping the Web, start calling it
25mArchitecting AI in Angular: WebMCP, A2UI, and the Agent Experience
30mPractical steps to make your website agent-ready