Building MCP Apps: Branded Experiences Inside ChatGPT

The content breaks down into six modules. We start with the MCP Apps spec itself — how UI resources are declared on a tool, how the host renders them, and what the security model looks like. From there we move into the Apps SDK and OpenAI's UI kit (buttons, cards, input controls, layout primitives) so attendees can see how to match ChatGPT's container while still expressing a brand. The third module covers the JSON-RPC bridge: how the iframe talks back to the host, tool invocation, and state sync.

Then we get into building and wiring up an MCP server that exposes UI resources, before doing a full live build of a branded app inside ChatGPT — covering layout, theming hooks, interaction patterns, and graceful fallbacks across clients (Claude, VS Code, Goose). We close with a deployment and developer-mode deep dive: registering apps, testing in ChatGPT developer mode, and shipping one experience that works across every MCP-compatible client without client-specific code. 

Attendees need to be comfortable with TypeScript basics, React fundamentals, and async/await. Everything MCP- and Apps-SDK-specific is covered from the ground up.