Hey, We've just open-sourced MCP-UI, a project that provides lean client and server SDKs to enable passing rich, interactive UI components over vanilla MCP. We believe it'll enable a future where AI interaction will be based on UI snippets for interactive experiences.

MCP-UI is meant to be a playground to bootstrap the initiative. Hopefully, through experimentation, the community can find the best way to enable this new future.

Under the hood, the SDK enables any MCP server to respond with an EmbeddedResource with a "ui://" URI. The client SDK allows hosts to render it using a supported method (currently raw HTML or an external app) and handles follow-up interactions through events (e.g., further tool calls). We need better delivery methods, starting with React Server Components and Remote-DOM for better security. There are many other things to accomplish, such as scalable protocol negotiation.

You can build with it today -

GitHub: https://github.com/idosal/mcp-ui (see demo video)

Docs: https://idosal.github.io/mcp-ui

UI-enabled chat client with compliant sample server: https://scira-mcp-chat-indol.vercel.app (type "show tasks" to see it in action!)

Inspector for local development: https://github.com/idosal/ui-inspector

Further reading on UI in MCP: https://github.com/orgs/modelcontextprotocol/discussions/287

Thoughts, changes, and contributions are more than welcome! Can't wait to see what we can build with it.