A lightweight, native desktop hub that manages Model Context Protocol (MCP) tools for AI clients, with dynamic tool discovery and profile-based identity isolation.

As Model Context Protocol (MCP) servers proliferate, developers face context bloat, configuration chaos, and security leaks when switching between accounts and AI clients such as Cursor, Claude Desktop, and VS Code.

Challenge

Connecting dozens of tools floods LLMs with unused definitions, consuming context windows and burning tokens. Switching between personal and work accounts means manually swapping API keys and JSON configs across many files, with no isolation or audit trail.

Solution

We designed and built a native, lightweight desktop hub:

  • Tauri & Rust Desktop Shell: A cross-platform system-tray app with a native UI and secure OS keychain integration.
  • Go & SSE Gateway: A local server handling JSON-RPC, managing tool lifecycles, and starting tools on demand in under 10ms.
  • Dynamic Tool Discovery: Exposes only two primordial meta-tools to AI clients, letting the model discover and activate tools on demand to keep the context window lean.
  • Profile Isolation: Complete separation of credentials, environment variables, and allowed tools between profiles (e.g., Work vs. Personal).

Results

The app runs in under 50MB of RAM and starts tools almost instantly. It eliminated configuration chaos for developers, sharply reduced LLM context consumption, and enforced security through strict profile isolation.