MCP Server Deepseek

🚀 mcp-server-deepseek Service

This is a server for the Model Context Protocol (MCP), providing access to the inference capabilities of DeepSeek-R1. It enables models without inference capabilities to generate better responses and improve performance through enhanced thinking.

• Read R1 + Sonnet Sets SOTA on Aider's Polyglot Benchmark

🚀 Quick Start

This server acts as a bridge between LLM applications and the inference capabilities of DeepSeek. It exposes the inference content of DeepSeek-R1 through MCP tools and can be used by any MCP-compatible client.

This server is particularly suitable for:

• Enhancing the response quality of models without native inference capabilities
• Accessing the thinking process of DeepSeek-R1 to solve complex problems
• Adding structured reasoning to Claude or other MCP-supported LLMs

✨ Features

• Access to DeepSeek-R1: Connect to DeepSeek's API and leverage the capabilities of its inference model.
• Structured Thinking: Return inference content in the <thinking> format.
• Integration with MCP: Fully compatible with the Model Context Protocol.
• Error Handling: Robust error handling and detailed logging.

📦 Installation

Prerequisites

• Python 3.13 or higher
• DeepSeek API key

Setup Steps

1. Clone the repository:

   git clone https://github.com/yourusername/mcp-server-deepseek.git
   cd mcp-server-deepseek
   

2. Create a virtual environment:

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate
   

3. Install the package:

   pip install -e .
   

4. Create a .env file to store DeepSeek's API credentials:

   cp .env.example .env
   

5. Edit the .env file and enter your API key and model details:

   MCP_SERVER_DEEPSEEK_MODEL_NAME=deepseek-reasoner
   MCP_SERVER_DEEPSEEK_API_KEY=your_api_key_here
   MCP_SERVER_DEEPSEEK_API_BASE_URL=https://api.deepseek.com
   

💻 Usage Examples

Running the Server

You can run the server directly:

mcp-server-deepseek

Or use the development mode in combination with MCP Inspector:

make dev

MCP Tools

The server provides a tool:

think_with_deepseek_r1

This tool sends a prompt to DeepSeek-R1 and returns its inference content.

Parameters:

• prompt (string): The complete user prompt to be processed.

Returns:

• A string containing the inference content of DeepSeek-R1 wrapped in <thinking> tags.

Basic Usage

When used with Claude or other MCP-supported LLMs, you can trigger the inference process by calling the tool:

Please use the think_with_deepseek_r1 tool and provide the following prompt:
"How to optimize neural networks for time series prediction?"

🔧 Technical Details

Testing

Commands for development and testing:

npx @modelcontextprotocol/inspector uv run mcp-server-deepseek

Logging

Logs are stored in the ~/.cache/mcp-server-deepseek/server.log file.

You can configure the log level through the environment variable LOG_LEVEL (default is DEBUG).

Troubleshooting

Common Issues

• API Key Issue: Ensure that the DeepSeek API key is correctly added to the .env file.
• Model Loading Failure: Check if the model name is correctly configured and confirm that the API service is running normally.

📚 Documentation

For the complete project documentation, please refer to GitHub Wiki