RAT Retrieval Augmented Thinking MCP

🚀 Deepseek-Thinking-Claude-3.5-Sonnet-CLINE-MCP Server

A Model Context Protocol (MCP) server that combines the inference capabilities of DeepSeek R1 with the answer generation capabilities of Claude 3.5 Sonnet through OpenRouter's unified API. This implementation uses a two-stage process: DeepSeek performs preliminary inference (with a 50k character context), and Claude is responsible for the final answer (with a 600k character context). Both are accessed via OpenRouter, and the inference results from DeepSeek are injected into Claude's answer.

🚀 Quick Start

This server integrates the inference ability of DeepSeek R1 and the answer generation ability of Claude 3.5 Sonnet. It uses a two - stage process to provide high - quality responses.

✨ Features

• Combines DeepSeek R1's inference and Claude 3.5 Sonnet's answer generation.
• Utilizes a two - stage process for better performance.
• Accesses both models through OpenRouter's unified API.

📦 Installation

Install using Smithery

1. Open your terminal.
2. Run the command:

npx smithery install @newideas99/Deepseek-Thinking-Claude-3.5-Sonnet-CLINE-MCP

Manual Installation

1. Clone the repository:

   git clone https://github.com/deepseek-clause/thinking-claude-sonnet-mcp.git
   cd thinking-claude-sonnet-mcp
   

2. Install dependencies:

   npm install
   

3. Create a .env file and add the following content:

   REASONING_API_KEY=your_reasoning_api_key_here
   CLAUSERSONNET_API_KEY=your_clauser_sonnet_api_key_here
   

💻 Usage Examples

Initialize the Server

node index.js

Tool Call Examples

1. Call the Inference Tool (DeepSeek R1)

const result = await use_mcp_tool({
  server_name: "deepseek-claude",
  tool_name: "generate_response",
  arguments: {
    prompt: "What is quantum computing?",
    showReasoning: true
  }
});

2. Get the Inference Result

const taskId = JSON.parse(result.content[0].text).taskId;
const statusResult = await use_mcp_tool({
  server_name: "deepseek-claude", 
  tool_name: "check_response_status",
  arguments: { taskId }
});

📚 Documentation

Automatically Rebuild the Development Environment

npm run watch

⚠️ Important Note

• Ensure that valid reasoning API keys and Clauser Sonnet API keys are configured.
• After installation, initialize the server before calling the tools.

🔧 Technical Details

Inference Stage (DeepSeek R1)

• Use OpenRouter's inference token feature.
• Modify the prompt to output 'completed' and capture the inference content.
• Extract the inference result from the response metadata.

Answer Stage (Claude 3.5 Sonnet)

• Receive the original prompt and the inference content from DeepSeek.
• Generate the final answer by combining context information and maintaining the conversation history.

📄 License

MIT License - See the LICENSE file for details.

Acknowledgments

Based on the RAT (Reinforced Retrieval Thinking) concept proposed by Skirano, which enhances AI response effects through structured reasoning and knowledge retrieval.