MCP-server-Qwen_Max

A Model Context Protocol (MCP) server implementation for the Qwen Max language model.

Why Node.js? This implementation uses Node.js/TypeScript as it currently provides the most stable and reliable integration with MCP servers compared to other languages like Python. The Node.js SDK for MCP offers better type safety, error handling, and compatibility with Claude Desktop.

• Node.js (v18 or higher)
• npm
• Claude Desktop
• Dashscope API key

To install Qwen Max MCP Server for Claude Desktop automatically via Smithery:

npx -y @smithery/cli install @66julienmartin/mcp-server-qwen_max --client claude

git clone https://github.com/66julienmartin/mcp-server-qwen-max.git
cd Qwen_Max
npm install

By default, this server uses the Qwen-Max model. The Qwen series offers several commercial models with different capabilities:

Provides the best inference performance, especially for complex and multi-step tasks.

Context window: 32,768 tokens

• Max input: 30,720 tokens
• Max output: 8,192 tokens
• Pricing: $0.0016/1K tokens (input), $0.0064/1K tokens (output)
• Free quota: 1 million tokens

Available versions:

• qwen-max (Stable)
• qwen-max-latest (Latest)
• qwen-max-2025-01-25 (Snapshot, also known as qwen-max-0125 or Qwen2.5-Max)

Balanced combination of performance, speed, and cost, ideal for moderately complex tasks.

Context window: 131,072 tokens

• Max input: 129,024 tokens
• Max output: 8,192 tokens
• Pricing: $0.0004/1K tokens (input), $0.0012/1K tokens (output)
• Free quota: 1 million tokens

Available versions:

• qwen-plus (Stable)
• qwen-plus-latest (Latest)
• qwen-plus-2025-01-25 (Snapshot, also known as qwen-plus-0125)

Fast speed and low cost, suitable for simple tasks.

• Context window: 1,000,000 tokens
• Max input: 1,000,000 tokens
• Max output: 8,192 tokens
• Pricing: $0.00005/1K tokens (input), $0.0002/1K tokens (output)
• Free quota: 1 million tokens

Available versions:

• qwen-turbo (Stable)
• qwen-turbo-latest (Latest)
• qwen-turbo-2024-11-01 (Snapshot, also known as qwen-turbo-1101)

To modify the model, update the model name in src/index.ts:

// For Qwen-Max (default)
model: "qwen-max"

// For Qwen-Plus
model: "qwen-plus"

// For Qwen-Turbo
model: "qwen-turbo"

For more detailed information about available models, visit the Alibaba Cloud Model Documentation https://www.alibabacloud.com/help/en/model-studio/getting-started/models?spm=a3c0i.23458820.2359477120.1.446c7d3f9LT0FY.

qwen-max-mcp/
├── src/
│   ├── index.ts             # Main server implementation
├── build/                   # Compiled files
│   ├── index.js
├── LICENSE
├── README.md
├── package.json
├── package-lock.json
└── tsconfig.json

1. Create a .env file in the project root:

DASHSCOPE_API_KEY=your-api-key-here

1. Update Claude Desktop configuration:

{
  "mcpServers": {
    "qwen_max": {
      "command": "node",
      "args": ["/path/to/Qwen_Max/build/index.js"],
      "env": {
        "DASHSCOPE_API_KEY": "your-api-key-here"
      }
    }
  }
}

npm run dev     # Watch mode
npm run build   # Build
npm run start   # Start server

• Text generation with Qwen models
• Configurable parameters (max_tokens, temperature)
• Error handling
• MCP protocol support
• Claude Desktop integration
• Support for all Qwen commercial models (Max, Plus, Turbo)
• Extensive token context windows

// Example tool call
{
  "name": "qwen_max",
  "arguments": {
    "prompt": "Your prompt here",
    "max_tokens": 8192,
    "temperature": 0.7
  }
}

The temperature parameter controls the randomness of the model's output:

Lower values (0.0-0.7): More focused and deterministic outputs Higher values (0.7-1.0): More creative and varied outputs

Recommended temperature settings by task:

Code generation: 0.0-0.3 Technical writing: 0.3-0.5 General tasks: 0.7 (default) Creative writing: 0.8-1.0

The server provides detailed error messages for common issues:

API authentication errors Invalid parameters Rate limiting Network issues Token limit exceeded Model availability issues

Contributions are welcome! Please feel free to submit a Pull Request.

MIT