io.github.Lohithry/divlens

What is DivLens MCP?

DivLens MCP is a high-performance Model Context Protocol (MCP) server written in Rust.

It bridges the gap between AI assistants and your machine — giving Claude, Cursor, Windsurf, and any other MCP-compatible agent live, structured access to hardware sensors, storage metrics, network diagnostics, process trees, developer runtimes, system logs, and more.

No cloud. No API keys. No configuration required. Just build and run.

"Why is my Mac slow?" → Claude calls get_live_metrics() → Instant answer.
"Is my SSD healthy?"  → Claude calls get_hardware_diagnostics() → SMART data returned.
"What's eating disk?"  → Claude calls get_advanced_storage_stats() → Largest files listed.

✦ 17 Diagnostic Tools

🚀 Install — One Command, Any Platform

No Rust required. No compilation. No manual config editing. The installer downloads a pre-built binary and automatically configures your AI clients.

macOS & Linux

curl -fsSL https://raw.githubusercontent.com/Lohithry/divlens-mcp/main/install.sh | bash

Windows (PowerShell — no admin required)

irm https://raw.githubusercontent.com/Lohithry/divlens-mcp/main/install.ps1 | iex

The installer will:

• ✅ Detect your OS and chip (Apple Silicon / Intel / Linux / Windows)
• ✅ Download the correct pre-built binary from GitHub Releases
• ✅ Verify the SHA-256 checksum
• ✅ Install to your PATH with no admin rights needed
• ✅ Auto-configure Claude Desktop, Cursor, Windsurf, and Antigravity
• ✅ Test the server works before finishing

Then just restart your AI client and ask "What's using my CPU right now?"

Build from Source (Advanced)

Requires Rust 1.82+.

git clone https://github.com/Lohithry/divlens-mcp.git
cd divlens-mcp/apps/core
cargo build --release
./target/release/divlens-core --mcp

Connect to Your AI

Claude Desktop

Config file: ~/Library/Application Support/Claude/claude_desktop_config.json (macOS)
or %APPDATA%\Claude\claude_desktop_config.json (Windows)

{
  "mcpServers": {
    "divlens": {
      "command": "/usr/local/bin/divlens-core",
      "args": ["--mcp"]
    }
  }
}

Quit and relaunch Claude Desktop. A 🔌 plug icon confirms the connection.

Cursor

Config file: ~/.cursor/mcp.json

{
  "mcpServers": {
    "divlens": {
      "command": "/usr/local/bin/divlens-core",
      "args": ["--mcp"]
    }
  }
}

Cmd+Shift+P → Reload Window

Windsurf

Config file: ~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "divlens": {
      "command": "/usr/local/bin/divlens-core",
      "args": ["--mcp"]
    }
  }
}

For complete setup details, see DEPLOYMENT.md.

How It Works

  ┌─────────────────────────────────────────┐
  │   AI Client  (Claude / Cursor / etc.)   │
  │         LLM reasoning lives here        │
  └──────────────────┬──────────────────────┘
                     │  JSON-RPC 2.0  (stdio)
                     ▼
  ┌─────────────────────────────────────────┐
  │          divlens-core  (Rust)           │
  │                                         │
  │  ┌───────────────┐  ┌───────────────┐   │
  │  │  MCP Layer    │  │  17 Tools     │   │
  │  │  (JSON-RPC)   │  │  (Rust + OS)  │   │
  │  └───────────────┘  └───────────────┘   │
  │  ┌───────────────┐  ┌───────────────┐   │
  │  │  SQLite Cache │  │  Native APIs  │   │
  │  │  (sysinfo/OS) │  │  (IOKit/WMI)  │   │
  │  └───────────────┘  └───────────────┘   │
  └─────────────────────────────────────────┘

      Zero cloud.  Zero API keys.  100% local.

Transport: Every MCP message is a newline-delimited JSON-RPC 2.0 object over stdio.
AI logic: DivLens never runs LLM inference — it only collects and returns raw system data.
Privacy: All data stays on your machine. Nothing is sent anywhere.

Project Structure

divlens-mcp/
└── apps/
    └── core/                      # Rust MCP engine
        ├── src/
        │   ├── tools/             # 17 tool implementations
        │   ├── mcp/               # JSON-RPC 2.0 protocol handler
        │   ├── mcp_server.rs      # stdio transport loop
        │   ├── collectors/        # Native OS data collectors
        │   │   ├── volatile/      # CPU, RAM, network (live)
        │   │   ├── persistent/    # Storage, hardware (cached)
        │   │   └── ondemand/      # Drivers, logs, packages
        │   ├── modules/           # Core business logic
        │   ├── db/                # SQLite caching layer
        │   ├── models/            # Shared data types
        │   └── utils/             # Shell env rehydration
        ├── Cargo.toml
        └── env.example

Optional: Semantic Memory

Enable the vector-memory feature to give recall_memory true semantic search using a local ONNX embedding model (no cloud, no API key):

cargo build --release --features vector-memory

When enabled, DivLens creates a local LanceDB vector store and uses fastembed to embed and recall past diagnoses semantically.

When disabled (default), recall_memory returns an empty list — no functionality is broken.

Verify the Server

Test the MCP wire protocol without a client:

# Initialize handshake
echo '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","clientInfo":{"name":"test","version":"0.1"}}}' \
  | divlens-core --mcp

# Call a tool directly
echo '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"get_live_metrics","arguments":{}}}' \
  | divlens-core --mcp

License

Licensed under the Apache License, Version 2.0.
See LICENSE for the full text.

Copyright © 2024 DivLens Contributors.