AI-QL/tuui

Download the app from the GitHub Releases page (Windows / Linux / macOS)
Configure LLM backends in src/main/assets/config/llm.json (or resources/assets/config/llm.json in built releases); MCP servers in mcp.json

Usage

• •Set up one or more LLM backends (JSON object for single chatbot or JSON array for multiple) — e.g. Qwen, OpenRouter/proxy, DeepInfra — each requiring name, apiKey, url, path, model, and mcp: true
• •Configure MCP servers in mcp.json; for remote servers use mcp-remote (npx -y mcp-remote https://domain/sse)
• •Config is stored in localStorage; clear it via Tray Menu → Clear Storage

Key features

• ✓MCP client support for Tools, Prompts, Resources, Sampling, and Elicitation (client-side); MCP Registry for real-time MCP server discovery
• ✓.mcpb bundle support (Model Context Protocol Bundles); MCP tool-call tracing; agent mode with specified tool selection
• ✓Cross-vendor LLM orchestration with dynamic configuration; automated application testing support; multilingual UI; TypeScript
• ✓Native devtools; wiki/docs at TUUI.com; DeepWiki integration for asking AI about the project

Best for

A local, account-free desktop chat client for people who want to orchestrate multiple LLM vendors and use MCP servers in one UI.

Caveats

• ⚠Requires an LLM backend that supports tool/function calling; npx-based servers need Node.js, uv/uvx servers need Python + UV, Docker-based servers need Docker
• ⚠macOS MCP config issues (ENOENT spawn, nvm interference) are documented as unresolved on some platforms; may need absolute paths or the workaround in issue #2
• ⚠90-second connection timeout can trigger on first launch while uv/uvx/npx runtimes are installed — pre-install packages if repo access is slow
• ⚠README notes the project is partly AI-generated and enforces strict lint/syntax checks for contributors