Model Context Protocol (MCP) server that integrates powerful DeepSeek language models (including the R1 reasoning model) with MCP-compatible applications, with support for anonymous proxy and multi-turn conversations.

Installation

Smithery

npx -y @smithery/cli install @dmontgomery40/deepseek-mcp-server --client claude

NPM Global

npm install -g deepseek-mcp-server

NPX Direct

npx -y deepseek-mcp-server

Configuration

Claude Desktop

{
  "mcpServers": {
    "deepseek": {
      "command": "npx",
      "args": [
        "-y",
        "deepseek-mcp-server"
      ],
      "env": {
        "DEEPSEEK_API_KEY": "your-api-key"
      }
    }
  }
}

Features

• Anonymous proxy usage for accessing DeepSeek API
• Automatic model switching from R1 to V3 when needed
• Multi-turn conversation support with full message history
• Resource discovery for available models and configurations
• Custom model selection (deepseek-reasoner/deepseek-chat)
• Temperature control (0.0 - 2.0)
• Maximum token limit configuration
• Top P sampling (0.0 - 1.0)
• Presence penalty (-2.0 - 2.0)
• Frequency penalty (-2.0 - 2.0)

Environment Variables

Required

• DEEPSEEK_API_KEY - Your DeepSeek API key for authentication

Usage Examples

What models are available?

What configuration options do I have?

What is the current temperature setting?

Start a multi-turn conversation. With the following settings: model: 'deepseek-chat', make it not too creative, and allow 8000 tokens.

use deepseek-reasoner

Notes

The server uses DeepSeek's R1 model (deepseek-reasoner) by default for state-of-the-art reasoning performance. V3 (deepseek-chat) is recommended for general use due to its speed and token efficiency. Multi-turn conversation support is useful for training/fine-tuning open source models and complex interactions that require context preservation. Can be tested locally using the MCP Inspector tool.