Back to Podcast Digest
AI Engineer16m

Building Interactive UIs in VS Code with MCP Apps — Marlene Mhangami & Liam Hampton, GitHub

TL;DR

  • MCP apps fix the "text-only" problem: Marlene contrasts early MCP outputs like ASCII art and emoji-heavy readmes with newer apps that return rich, interactive components directly inside chat.

  • VS Code is the host, Copilot is the client, and the server provides the capability: The talk clearly separates MCP's core pieces and recommends finding servers through VS Code's @mcp extension list or GitHub rather than installing random ones from the internet.

  • The UI flow is server to resource reference to sandboxed iframe: An MCP server returns tool results plus a reference to HTML, then VS Code fetches and renders that UI in an iframe where users can interact safely.

  • Interactive chat matters most when clicking beats prompting: Marlene uses data exploration and in-chat commerce as examples where users should inspect charts or complete checkout flows without constantly typing follow-up instructions.

  • Companies like Shopify, Excalidraw, and Figma are already building this way: Shopify is focused on preserving brand identity inside chat, while Excalidraw and Figma use MCP apps for editable diagrams and generated components.

  • Liam's flame graph demo shows the practical value: He profiles a Go app over 5 seconds via a local MCP server and renders the results as an interactive React UI in Copilot, replacing back-and-forth questions like "is this good or bad?" with a visual answer.

The Breakdown

A live VS Code demo shows MCP apps turning plain tool output into interactive UI inside chat, including a flame graph profiler that renders live in an iframe instead of dumping a wall of JSON. Marlene Mhangami and Liam Hampton make the case that this is how agents stop feeling like glorified link generators and start keeping users inside the workflow.

Was This Useful?

Share