RAT (Retrieval Augmented Thinking)

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

A Model Context Protocol (MCP) server that combines the reasoning ability of Deepseek R1 and the sonnet response generation ability of Claude 3.5, implemented via OpenRouter. This implementation adopts a two - stage process where Deepseek provides structured reasoning, which is then integrated into Claude's response generation.

✨ Features

• Two - stage processing:

  • Use Deepseek R1 for initial reasoning (50k character context).
  • Use Claude 3.5 Sonnet for final response (600k character context).
  • Both models are accessed through OpenRouter's unified API.
  • Inject Deepseek's reasoning tokens into Claude's context.

• Intelligent conversation management:

  • Detect active conversations using file modification time.
  • Handle multiple concurrent conversations.
  • Automatically filter ended conversations.
  • Support context clearing when needed.

• Optimized parameters:

  • Model - specific context limits:
    • Deepseek: 50,000 characters for focused reasoning.
    • Claude: 600,000 characters for comprehensive responses.
  • Recommended settings:
    • temperature: 0.7 (balance of creativity).
    • top_p: 1.0 (full probability distribution).
    • repetition_penalty: 1.0 (prevent repetition).

📦 Installation

Install using Smithery

Automatically install Deepseek - Thinking - Claude 3.5 Sonnet for Claude Desktop via Smithery:

npx -y @smithery/cli install @newideas99/Deepseek-Thinking-Claude-3.5-Sonnet-CLINE-MCP --client claude

Manual installation

1. Clone the repository:

git clone https://github.com/yourusername/Deepseek-Thinking-Claude-3.5-Sonnet-CLINE-MCP.git
cd Deepseek-Thinking-Claude-3.5-Sonnet-CLINE-MCP

2. Create and activate a virtual environment (optional but recommended):

python -m venv env
source env/bin/activate  # On Linux/Mac
env\Scripts\activate      # On Windows

3. Install dependencies:

pip install -r requirements.txt

4. Run the server:

python app.py

💻 Usage Examples

Basic Usage

• Reasoning stage (Deepseek R1):

  • Use OpenRouter's reasoning token function.
  • Modify the prompt to output 'completed' while capturing the reasoning content.
  • Extract the reasoning from the response metadata.

• Response stage (Claude 3.5 Sonnet):

  • Receive the original prompt and Deepseek's reasoning content.
  • Generate the final response, integrating the reasoning content.
  • Maintain the conversation context and history.

🔧 Development

For feature development with automatic reconstruction:

npm run watch

📄 License

MIT License - See the LICENSE file for details.

🔗 Contributors

Based on Skirano's concept RAT (Retrieval - Augmented Thinking), which enhances AI responses through structured reasoning and knowledge retrieval.