A chat UI that treats LLMs and tool servers as hot-swappable parts

What it does This project is essentially polished plumbing: a Next.js interface that wires Vercel’s AI SDK to Model Context Protocol servers so you don’t have to. You connect to remote tool providers over HTTP or SSE, pick an LLM backend, and chat with the augmented result. The UI is built with shadcn/ui and Tailwind, and reasoning models are supported.

The interesting bit The app treats LLMs and tool servers as interchangeable commodities rather than locked-in integrations. That sounds obvious, but it saves you from rewriting the same chat scaffolding every time a new model or MCP server drops.

Key highlights

• Swaps between AI providers via the Vercel AI SDK, which abstracts streaming and tool-call wiring.
• Connects to any MCP-compatible server over HTTP or SSE, including Composio, Zapier, and Hugging Face.
• Supports reasoning models and streaming responses.
• Uses Next.js App Router and shadcn/ui components.

Caveats

• Transport is limited to HTTP and SSE; stdio-based MCP servers aren’t mentioned.
• The README promises “built-in tool integration” but never lists what those tools actually are.

Verdict Grab this if you need a drop-in chat UI for MCP servers and want to avoid boilerplate wiring. Skip it if you’re looking for a novel backend or deep customization beyond the provided settings panel.