Deepseek Claude

You run a two-stage MCP server that first lets DeepSeek R1 perform structured reasoning and then passes that reasoning to Claude 3.5 Sonnet for final response generation. This setup leverages a unified API to produce thoughtful, context-aware answers while keeping conversations organized and scalable across multiple sessions.

How to use

Use this MCP server with an MCP client to generate responses that blend DeepSeek’s reasoning with Claude’s fluent final output. When you start a new conversation, the server processes your prompt through the two stages and returns a task-based result. You can optionally request the reasoning to be shown and control whether to clear the conversation context between turns. The system maintains context for ongoing conversations and can filter out ended chats automatically.

How to install

Prerequisites: you need Node.js installed on your machine to run the MCP server locally. You also require an OpenRouter API key to access both DeepSeek and Claude models.

Step-by-step commands to set up the MCP server locally.

# Install via Smithery (automatic setup)

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

# Manual installation

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

npm install

# Environment setup
# 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

# Build the server
npm run build

# Start the server (example)
npm run start

Note: The exact commands to start the server may vary slightly depending on your environment. Follow the provided scripts in your project to run the built server.

## Additional sections

Configuration and usage details are focused on getting you a working MCP server quickly. The key pieces you need to know are the MCP command to run locally, the required environment variables, and how to integrate with your MCP client or Cline-based workflow.

Security and key management: store your OpenRouter API key securely in your environment. Do not commit keys to version control. Rotate keys periodically and use separate keys for development and production environments.

Notes on operation: the server uses a polling mechanism for long-running requests. A request returns a task ID immediately, and you poll for status until the final response is available. If you request reasoning, DeepSeek’s reasoning tokens will be included in the final result metadata.

Troubleshooting tips: ensure your API key is valid and accessible from your runtime environment, verify the build output path matches your start command, and check that your MCP client is configured to point to the correct server instance and port.

## Server connection details

The MCP server can be run locally using a standard runtime command or connected via a remote MCP URL if provided. In this setup, you run the server as a local process and point your MCP client to that process.

## Tools and endpoints

The server exposes two primary tools for interacting with conversations and responses. The first generates a response to a prompt and can optionally show the reasoning. The second checks the status of an ongoing generation task.

## Available tools

### generate\_response

Main tool to generate a response for a given prompt, with optional reasoning display, context clearing, and history inclusion.

### check\_response\_status

Tool to poll the status of an ongoing generation task using the taskId returned by generate\_response.