nyvyn/lmstudio-mcp

LM Studio MCP Server

This IS a local MCP server that bridges LM Studio with MCP-compatible clients like Claude Code. When you build and run this project, it creates a running MCP server that exposes LM Studio's local API through standardized MCP tools.

⚡ Quick Start with Claude Code

1. Add This Server to Claude Code

In your project directory where you want to use LM Studio, add this MCP server directly from GitHub:

claude mcp add lmstudio-mcp -- npx github:nyvyn/lmstudio-mcp

Or if you need to specify a custom base URL:

claude mcp add lmstudio-mcp -- npx github:nyvyn/lmstudio-mcp --base-url http://localhost:1234

You can verify it was added with:

claude mcp list

2. Start LM Studio

1. Launch LM Studio
2. Load a model
3. Start the local server (should be accessible at http://localhost:1234)

3. Use in Claude Code

Once configured, you can interact with your local LM Studio models:

• "Check if LM Studio is running"
• "What models do I have available?"
• "Generate a story about dragons using my local model"
• "Use temperature 0.9 to write a creative poem"

🚀 Features

• Health Check: Verify LM Studio API connectivity
• Model Management: List available models and identify the currently loaded model
• Chat Completions: Generate text completions with configurable parameters
• System Integration: Access system information and basic utilities
• Robust Error Handling: Clear error messages and graceful failure handling

📋 Requirements

• Node.js 18.0+
• LM Studio running locally with API enabled (default: http://localhost:1234)
• Claude Code or other MCP-compatible client

🛠️ Installation & Setup

1. Clone and Install Dependencies

git clone <repository-url>
cd lmstudio-mcp
npm install

2. Build the Server

npm run build

🔧 Available Tools

LM Studio Integration Tools

• lmstudio_health_check - Check if LM Studio API is running and accessible
• lmstudio_list_models - Get a list of all available models in LM Studio
• lmstudio_get_current_model - Identify the currently loaded model
• lmstudio_chat_completion - Generate text completions with configurable:
  • Custom prompts and system prompts
  • Temperature control (0-2)
  • Max token limits (1-4096)

Utility Tools

• echo - Echo back provided text

🏃‍♂️ Development

Development Mode

npm run dev

Production Build

npm run build
npm start

Quick Start Script

./start-server.sh

Linting & Type Checking

npm run lint
npm run typecheck

🔍 Testing

Test with Claude Code

# List your MCP servers
claude mcp list

# Remove if needed
claude mcp remove lmstudio-mcp

Test the server directly with JSON-RPC:

# List available tools
echo '{"jsonrpc": "2.0", "id": 1, "method": "tools/list", "params": {}}' | node dist/index.js

# Initialize connection
echo '{"jsonrpc": "2.0", "id": 1, "method": "initialize", "params": {"protocolVersion": "2024-11-05", "capabilities": {}, "clientInfo": {"name": "test-client", "version": "1.0.0"}}}' | node dist/index.js

🛠️ Configuration

The server connects to LM Studio at http://localhost:1234 by default. You can configure a custom base URL using command line arguments:

Command Line Arguments

npx tsx src/index.ts --base-url http://localhost:1234

Note: The server automatically appends /v1 to the base URL for API requests.

Claude Code MCP Configuration

Add to your Claude Code MCP configuration:

{
  "mcpServers": {
    "lmstudio": {
      "command": "npx",
      "args": ["tsx", "./src/index.ts", "--base-url", "http://localhost:1234"]
    }
  }
}

Examples

# Connect to LM Studio on different port
npx tsx src/index.ts --base-url http://localhost:8080

# Connect to remote LM Studio instance
npx tsx src/index.ts --base-url http://192.168.1.100:1234