OwnRig MCP Server

AI hardware compatibility data for any MCP-compatible assistant. Query 50 models, 25 devices, 14 builds, 9 ready-to-buy machines, and 663 compatibility entries.

Transport: stdio

Install

npm install -g ownrig-mcp

Or use directly with npx:

npx ownrig-mcp

Tools

| Tool | Description | |------|-------------| | query_model | Get details for a specific AI model (VRAM, formats, use cases) | | query_device | Get specs for a GPU or Apple Silicon device | | query_compatibility | Check if a model runs on a device (tokens/sec, VRAM fit) | | list_models | List models with optional use_case / family filter | | list_devices | List devices with optional type / min_vram filter | | list_builds | List curated builds with optional tier / profile filter | | list_systems | List ready-to-buy machines (Mac, Dell, ASUS) with optional brand / type filter | | query_system | Get full details for a specific ready-to-buy machine | | recommend_build | Full recommendation engine — 3 paths (model→hw, workflow→hw, hw→models) | | find_models_for_device | "What can I run on my RTX 4090?" | | find_devices_for_model | "What GPU do I need for Llama 3.1 70B?" | | list_workflows | List workflow profiles (tools → hardware requirements) |

Usage with Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "ownrig": {
      "command": "npx",
      "args": ["-y", "ownrig-mcp"]
    }
  }
}

Usage with Cursor

Add to .cursor/mcp.json in your project:

{
  "mcpServers": {
    "ownrig": {
      "command": "npx",
      "args": ["-y", "ownrig-mcp"]
    }
  }
}

Usage from source (development)

If you have the OwnRig repo cloned:

# From project root
npm install
npm run generate:rec-data
npm run mcp

The mcp script builds a self-contained bundle via esbuild (resolving all @/ path aliases) then runs it. Running tsx mcp-server/index.ts directly does not work because the engine uses TypeScript path aliases that tsx cannot resolve transitively across module boundaries.

For your MCP client config, point to the built bundle:

{
  "mcpServers": {
    "ownrig": {
      "command": "node",
      "args": ["mcp-server/dist/index.mjs"],
      "cwd": "/path/to/ownrig"
    }
  }
}

Example queries

Once connected, ask your AI assistant:

• "What GPU do I need to run Llama 3.1 70B locally?"
• "Can an RTX 4090 run Qwen 3 32B?"
• "Recommend a build for running AI coding tools with Cursor"
• "What models can I run on my M4 Max MacBook Pro?"
• "Compare the Mac Studio M4 Ultra vs a custom build for AI"

Data

This package includes a snapshot of OwnRig's verified hardware compatibility data. The data is updated with each package release.

• 50 AI models with VRAM requirements per quantization level
• 25 GPUs and Apple Silicon devices with specs and pricing
• 14 curated PC builds with component lists and benchmarks
• 9 ready-to-buy machines (Mac, Dell, ASUS, NVIDIA)
• 663 model × device × quantization compatibility entries
• 7 workflow profiles mapping AI tools to hardware needs

Source: ownrig.com | License: MIT