A Model Context Protocol (MCP) server that combines DeepSeek R1's reasoning capabilities with Claude 3.5 Sonnet's response generation through OpenRouter. This implementation uses a two-stage process where DeepSeek provides structured reasoning which is then incorporated into Claude's response generation.

• Two-Stage Processing:

  • Uses DeepSeek R1 for initial reasoning (50k character context)
  • Uses Claude 3.5 Sonnet for final response (600k character context)
  • Both models accessed through OpenRouter's unified API
  • Injects DeepSeek's reasoning tokens into Claude's context

• Smart Conversation Management:

  • Detects active conversations using file modification times
  • Handles multiple concurrent conversations
  • Filters out ended conversations automatically
  • Supports 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 for balanced creativity
    • top_p: 1.0 for full probability distribution
    • repetition_penalty: 1.0 to prevent repetition

To install DeepSeek Thinking with Claude 3.5 Sonnet for Claude Desktop automatically via Smithery:

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

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. Install dependencies:

npm install

3. Create a .env file with your OpenRouter API key:

# Required: OpenRouter API key for both DeepSeek and Claude models
OPENROUTER_API_KEY=your_openrouter_api_key_here

# Optional: Model configuration (defaults shown below)
DEEPSEEK_MODEL=deepseek/deepseek-r1  # DeepSeek model for reasoning
CLAUDE_MODEL=anthropic/claude-3.5-sonnet:beta  # Claude model for responses

4. Build the server:

npm run build

Add to your Cline MCP settings (usually in ~/.vscode/globalStorage/saoudrizwan.claude-dev/settings/cline_mcp_settings.json):

{
  "mcpServers": {
    "deepseek-claude": {
      "command": "/path/to/node",
      "args": ["/path/to/Deepseek-Thinking-Claude-3.5-Sonnet-CLINE-MCP/build/index.js"],
      "env": {
        "OPENROUTER_API_KEY": "your_key_here"
      },
      "disabled": false,
      "autoApprove": []
    }
  }
}

The server provides two tools for generating and monitoring responses:

Main tool for generating responses with the following parameters:

{
  "prompt": string,           // Required: The question or prompt
  "showReasoning"?: boolean, // Optional: Show DeepSeek's reasoning process
  "clearContext"?: boolean,  // Optional: Clear conversation history
  "includeHistory"?: boolean // Optional: Include Cline conversation history
}

Tool for checking the status of a response generation task:

{
  "taskId": string  // Required: The task ID from generate_response
}

The server uses a polling mechanism to handle long-running requests:

1. Initial Request:

   • generate_response returns immediately with a task ID
   • Response format: {"taskId": "uuid-here"}

2. Status Checking:

   • Use check_response_status to poll the task status
   • Note: Responses can take up to 60 seconds to complete
   • Status progresses through: pending → reasoning → responding → complete

Example usage in Cline:

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

// Get taskId from result
const taskId = JSON.parse(result.content[0].text).taskId;

// Poll for status (may need multiple checks over ~60 seconds)
const status = await use_mcp_tool({
  server_name: "deepseek-claude",
  tool_name: "check_response_status",
  arguments: { taskId }
});

// Example status response when complete:
{
  "status": "complete",
  "reasoning": "...",  // If showReasoning was true
  "response": "..."    // The final response
}

For development with auto-rebuild:

npm run watch

1. Reasoning Stage (DeepSeek R1):

   • Uses OpenRouter's reasoning tokens feature
   • Prompt is modified to output 'done' while capturing reasoning
   • Reasoning is extracted from response metadata

2. Response Stage (Claude 3.5 Sonnet):

   • Receives the original prompt and DeepSeek's reasoning
   • Generates final response incorporating the reasoning
   • Maintains conversation context and history

MIT License - See LICENSE file for details.

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

This implementation specifically combines DeepSeek R1's reasoning capabilities with Claude 3.5 Sonnet's response generation through OpenRouter's unified API.