Two ways to ship interactive UI inside an AI chat went mainstream in 2026: MCP Apps, the open extension to Model Context Protocol, and the OpenAI Apps SDK, the framework for ChatGPT-native apps. They sound like rivals, but the relationship is stranger — and it decides how portable your work will be. Here's the split, and how to choose.
The key relationship
The OpenAI Apps SDK is built on MCP Apps. OpenAI's own docs tell developers to "build around the MCP Apps standard for portability, then layer on ChatGPT extensions where they improve the ChatGPT experience." So the two largest US AI labs now share a wire format: a tool or resource built against MCP Apps runs on ChatGPT, Claude, Gemini, Copilot and Cursor without per-vendor engineering.
MCP Apps — the open contract
MCP Apps is the official MCP extension that lets a server return rich UI — dashboards, forms, multi-step flows — rendered in a sandboxed iframe with a JSON-RPC bridge. It's vendor-neutral by design, and because it's just an extension of the protocol you already use for tools, an existing MCP server adopts it incrementally rather than as a rewrite.
OpenAI Apps SDK — the ChatGPT layer
The Apps SDK is the consumer-facing surface on top of that spec. It defines both the logic and the interface for apps that live inside ChatGPT conversations, and it adds ChatGPT-specific niceties. The trade is reach for lock-in: hundreds of millions of ChatGPT users, but a surface tuned to one client.
How to choose
Build the core against MCP Apps for portability, then add Apps SDK extensions only where they earn their keep inside ChatGPT. That asymmetry — OpenAI ships a vendor SDK on top of the open spec, Anthropic just ships the spec — is why MCP Apps is on track to be the default open contract for the rest of 2026. If you can only invest once, invest in the portable layer.
Going further
Start with what are MCP Apps and use an MCP server in ChatGPT Developer Mode. See how the protocol is evolving in the 2026-07-28 spec changes, and where Skills fit vs MCP.