DeepSeek MCP Server

An MCP server implementation that provides code generation and completion capabilities using the DeepSeek API, with support for tool chaining and cost optimization.

Features

• Code generation with language-specific support
• Code completion with context awareness
• Code optimization with multiple targets
• Tool chaining for complex operations
• Built-in caching for cost optimization
• TypeScript implementation with full type safety

Tools

1. generate_code

Generate code using DeepSeek API with language-specific support.

{
  "name": "generate_code",
  "params": {
    "prompt": "Write a function that sorts an array",
    "language": "typescript",
    "temperature": 0.7
  }
}

2. complete_code

Get intelligent code completions based on existing context.

{
  "name": "complete_code",
  "params": {
    "code": "function processData(data) {",
    "prompt": "Add input validation and error handling",
    "temperature": 0.7
  }
}

3. optimize_code

Optimize existing code for performance, memory usage, or readability.

{
  "name": "optimize_code",
  "params": {
    "code": "your code here",
    "target": "performance"
  }
}

4. execute_chain

Execute a chain of tools in sequence, with context passing between steps.

{
  "name": "execute_chain",
  "params": {
    "steps": [
      {
        "toolName": "generate_code",
        "params": {
          "prompt": "Create a REST API endpoint",
          "language": "typescript"
        }
      },
      {
        "toolName": "optimize_code",
        "params": {
          "target": "performance"
        }
      }
    ]
  }
}

Tool Chaining

Tool chaining allows you to combine multiple operations into a single workflow. Each tool in the chain can access the results of previous tools through the chain context.

Example chain:

1. Generate initial code
2. Complete the code with error handling
3. Optimize the final result

{
  "steps": [
    {
      "toolName": "generate_code",
      "params": {
        "prompt": "Create a user authentication function",
        "language": "typescript"
      }
    },
    {
      "toolName": "complete_code",
      "params": {
        "prompt": "Add input validation and error handling"
      }
    },
    {
      "toolName": "optimize_code",
      "params": {
        "target": "security"
      }
    }
  ]
}

Cost Optimization

The server implements several strategies to optimize API costs:

1. Request caching with TTL
2. Chain result caching
3. Smart prompt construction
4. Metadata tracking for usage analysis

Development

To start development:

npm run dev

To clean and rebuild:

npm run rebuild

Installation

1. Clone the repository
2. Install dependencies:

npm install

3. Build the project:

npm run build

4. Configure your DeepSeek API key in the MCP settings file:

{
  "mcpServers": {
    "deepseek": {
      "command": "node",
      "args": ["/path/to/deepseek-mcp/build/index.js"],
      "env": {
        "DEEPSEEK_API_KEY": "your-api-key"
      }
    }
  }
}

Usage

The server can be used with any MCP-compatible client. Here's an example using the MCP CLI:

mcp use deepseek generate_code --params '{"prompt": "Write a hello world program", "language": "python"}'

Requirements

• Node.js >= 18.0.0
• DeepSeek API key
• MCP-compatible client

License

ISC