When chatbots grow buttons: Building MCP apps
summary
FastMCP plus a new Python library called Prefab lets servers render full interactive UIs inside chat by returning a JSON-backed component instead of raw data; examples include an in-chat Excalidraw embed, sortable data tables paired with pie charts, and a file upload flow that avoids tokenizing file bytes. Prefab ships about 125 components, compiles Python to a React-renderable JSON protocol, adds a recent TypeScript port, and previews app-provided tools so agents can call UI actions directly.
AI-written summary · about
chapters (14)
- 0:00Introduction to MCP apps
- 1:02Traditional MCP tools limits
- 2:06What are MCP apps
- 3:51Frontend development challenge
- 4:54Prefab components for MCP
- 6:19UI with Prefab in Python
- 8:04FastMCP for MCP apps
- 9:27Complex UIs with Prefab
- 10:30Generative UI with MCP
- 10:50File uploads with MCP
- 11:53Customization in Prefab
- 12:36Prefab docs and playground
- 13:21Future of MCP tools
- 14:45Closing and call to action
related talks
22mMCP UI: Extending the frontier
38mEvolving AI chat with MCP Apps
26mBuild your First MCP App
29mMCP Apps: Primitives, discovery, and the Future of Software
16mBuilding Interactive UIs in VS Code with MCP Apps
23mIntroducing WebMCP: Agents in the Browser
21mBeyond Components: The Future of AI UI
20mLetting AI Interface with your App with MCP