MCP-WIP: integration proposal

[asterd]

Hi everyone,

A colleague and I have developed a small extension to the MCP protocol designed to orchestrate UI components within a chat environment.

The main idea is to enable an agent to select the most appropriate widget for a user's request, if needed, from a pool of strictly-formatted declarative widget UI manifests. This allows the frontend to instantiate the chosen widget from its manifest, along with any optional parameters. The manifests are defined as MCP resources, serving as a unique source of truth. The widgets themselves are technology-agnostic, with specific frontend framework implementations (React, Flutter, etc.) bridged via dedicated SDKs.

Each widget can be seen as a part of a traditional application. Our client manages switching between widgets, context management, MCP connection, tool calls, and LLM invocation. The frontend SDK handles injecting user interactions and intents from the widget into the context (and the reverse path will be supported as well), in a manner similar to your state management approach.

We also emphasize that, having approached the widget discovery system as a freely composable (RAG) system, there are no limits on the number of widgets that can be indexed, avoiding the performance limitations of current tools.

We are aware that the solution is still immature and that you already have integrations with libraries like ag-ui that implement similar solutions (but with fundamental differences).

What do you think?
Do you see opportunities for integration into your framework or extensions, also thanks to your contribution?

Here is the link to our project: https://github.com/Andurilx64/mcp-wip

[yuvarajsl]

Hey @asterd,
Thanks for reaching out and supporting Agno. I’ve shared this with the team, we’re working through all requests one by one and will get back to you soon.
If it’s urgent, please let us know. We appreciate your patience!