MCP Apps, the Model Context Protocol's first official extension, turns AI responses into interactive interfaces
Key Points
- With "MCP Apps," the Model Context Protocol (MCP) receives an official extension that enables AI tools to embed interactive interfaces directly into chats.
- Instead of requesting each interaction step via prompt for complex data such as database queries with hundreds of rows, users can now sort, filter, and view details themselves.
- Claude, Goose, and Visual Studio Code Insiders already support MCP Apps, with ChatGPT following this week.
The Model Context Protocol just got its first official extension with "MCP Apps." AI tools can now embed interactive interfaces directly into conversations.
MCP Apps is now available on Github, the MCP Core Maintainers announced. Tools can return dashboards, forms, visualizations, and multi-step workflows that render right in the chat.
MCP Apps tackles a fundamental problem: when a tool returns database queries with hundreds of rows, the model can summarize, but users often want to dig into the content themselves. With text-only responses, every interaction requires a new prompt.
The update fills this gap. The model stays in the loop and sees what the user is doing, while the UI handles what text alone can't: live updates, native media viewers, persistent states, and direct manipulation through a graphical interface.
The documentation lists use cases like configuration wizards, document review with highlighted PDF sections, and real-time server metrics monitoring. The ext-apps repository includes the SDK and working examples for 3D visualization, interactive maps, document display, and music notation.
Major platforms already on board
For the first time, developers can ship an interactive MCP experience that works across multiple clients without writing client-specific code, the MCP Core Maintainers claim. The extension builds on MCP-UI and the OpenAI-Apps SDK, and is the open standard built on MCP primitives, an Anthropic spokesperson told THE DECODER.
Claude already supports MCP Apps on web and desktop, along with Goose and Visual Studio Code Insiders. ChatGPT is expected to follow this week. JetBrains, AWS, and Google Deepmind have also signaled interest.
All UI content runs in sandboxed iframes with restricted permissions. Hosts can inspect HTML before rendering, communication flows through loggable JSON-RPC messages, and explicit user consent may be required for UI-initiated tool calls. Users should still vet servers carefully before connecting, the announcement notes.
AI News Without the Hype – Curated by Humans
As a THE DECODER subscriber, you get ad-free reading, our weekly AI newsletter, the exclusive "AI Radar" Frontier Report 6× per year, access to comments, and our complete archive.
Subscribe now