skrstop/MCPHub-Desktop

What it solves

MCPHub Desktop provides a local-first management system for the Model Context Protocol (MCP). It solves the fragmentation of managing multiple MCP servers, AI agent skills, and RAG knowledge bases by consolidating them into a single desktop application that runs entirely on the user's machine, ensuring privacy and offline availability.

How it works

Built with Tauri 2 and Rust, the application hosts MCP server processes locally and provides a unified interface for configuration and debugging. It implements a local RAG pipeline using GGUF embedding models (via the candle backend) and a vector database (LanceDB) to create "long-term memory" for AI agents. It also includes a skill management system that can deploy tools to various AI clients (like Cursor or Claude Code) via symlinks or file copies.

Who it’s for

Individual users and developers who use multiple AI agents and want a centralized, private way to manage their MCP servers, custom skills, and personal knowledge bases without relying on cloud services.

Highlights

• Local-First RAG: Automatic document chunking (using tree-sitter for code) and vectorization for cross-session agent memory.
• Skills Distribution: One-click installation and synchronization of AI skills across dozens of popular agents (e.g., Windsurf, Cline, Roo Code).
• Bundled Runtimes: Includes Node, UV, and Bun runtimes out-of-the-box, removing the need for manual environment setup.
• Embedded Model Support: Local inference for GGUF embedding models including Gemma3, Qwen3, and Modern-BERT.
• MCP Tool Integration: Exposes RAG capabilities (search, create, update) as MCP tools that agents can call directly.