mlx-mcp-server

MCP server bridging Claude Code and Claude Desktop to a locally-running LLM. Optimized for MLX LM on Apple Silicon, with support for any OpenAI-compatible backend (Ollama, LM Studio, etc.).

Install

pip install mlx-mcp-server

Setup

1. Start your local LLM

# Apple Silicon — MLX LM
mlx_lm.server --model mlx-community/Mistral-7B-Instruct-v0.3-4bit

# Intel Mac / Linux — Ollama
ollama serve && ollama pull mistral

2. Add to Claude Desktop

# MLX (Apple Silicon)
mlx-mcp-server install

# Ollama
mlx-mcp-server install --base-url http://localhost:11434 --model mistral

# Preview before writing
mlx-mcp-server install --dry-run

3. Add to Claude Code

mlx-mcp-server install --claude-code

Restart Claude Desktop / Claude Code after installing.

Tools

Configuration

Set via environment variables (or the install command sets them automatically):

Manual Claude Desktop config

If you prefer to edit the config file directly (~/Library/Application Support/Claude/claude_desktop_config.json on macOS):

{
  "mcpServers": {
    "mlx": {
      "command": "mlx-mcp-server",
      "env": {
        "MLX_BASE_URL": "http://localhost:8080",
        "MLX_DEFAULT_MODEL": ""
      }
    }
  }
}

Requirements

• Python 3.11+

• A running OpenAI-compatible LLM backend:

  • MLX LM (Apple Silicon, recommended)

  • Ollama (Intel Mac / Linux)

  • LM Studio or any other OpenAI-compatible server

License

MIT