webui: MCP client with low coupling to current codebase

[ServeurpersoCom]

Make sure to read the contributing guidelines before submitting a PR

• multi-transport MCP client
• full agentic orchestrator
• isolated, idempotent singleton initialization
• typed SSE client
• normalized tool-call accumulation pipeline
• integrated reasoning, timings, previews, and turn-limit handling
• complete UI section for MCP configuration
• dedicated controls for relevant parameters
• opt-in ChatService integration that does not interfere with existing flows

TODO: increase coupling with the UI for structured tool-call result rendering, including integrated display components and support for sending out-of-context images (persistence/storage still to be defined).

llama-webui-mcp-client.mp4

[allozaur]

@ServeurpersoCom this PR needs updating after #17470

[banyan-god]

Cant wait for this to be merged. In addition are there any other ways to add tool calls ? Maybe like how custom gpts do

[ServeurpersoCom]

To answer your question: MCP is flexible enough to cover pretty much any tool-calling scenario you can think of. You can build RAG frontends, agentic development sandboxes, query personal databases or local APIs, even hook up stable-diffusion.cpp behind an MCP server: the protocol doesn't care what's on the other side.

That said, this implementation is a pure client-side approach (browser). If you want to reach MCP servers beyond 127.0.0.1, you'll need either a backend proxy or a small home server setup to handle CORS.

As for the merge timeline: I need to do a few more code review passes myself and validate the architecture with @allozaur before this is ready. Patience :)

[jacekpoplawski]

Would it be a good idea to add some sample server to make it easy to start using the tools? (If I understand correctly, it could be a simple Python script?)

[ServeurpersoCom]

Yes, we'll be able to do lots of fun things, in different languages ​​including Python, and there are already plenty of ready-made MCP server examples on GitHub. Basically, anything that runs on 127.0.0.1 with Streamable-HTTP or WebSocket.
With a small Python script running locally, you can wire up pretty much anything an LLM might find useful