mcp-deepthinking

An MCP (Model Context Protocol) server that provides a deep thinking tool powered by Groq Large Language Models (LLMs). This server enables complex reasoning, multi-step problem solving, and advanced planning capabilities through a simple API interface.

Features

• Deep Thinking Tool: Exposes a tool named deepthinking for complex reasoning tasks.
• Groq LLM Integration: Utilizes Groq's high-performance language models via the Groq API.
• Configurable Models: Supports multiple Groq models, configurable via environment variables.
• Async & Streaming Support: Designed for efficient, asynchronous operation.
• Easy Integration: Compatible with any MCP client.

Requirements

• Python >= 3.12
• Groq API Key (sign up at https://console.groq.com/)
• Supported Groq models:
  • deepseek-r1-distill-llama-70b (default)
  • deepseek-r1-distill-qwen-32b
  • qwen-qwq-32b

Installation

You can install dependencies using either uv (recommended) or pip.

Using uv (recommended)

uv pip install -r requirements.txt
# or directly
uv pip install fastmcp>=0.4.1 langchain-groq>=0.3.2 mcp[cli]>=1.6.0 python-dotenv>=1.1.0

Using pip

pip install fastmcp>=0.4.1 langchain-groq>=0.3.2 mcp[cli]>=1.6.0 python-dotenv>=1.1.0

Configuration

Create a .env file in the project root with the following content:

GROQ_API_KEY=your_groq_api_key_here
MODEL_ID=deepseek-r1-distill-llama-70b  # optional, defaults to this model

• GROQ_API_KEY (required): Your Groq API key.
• MODEL_ID (optional): One of the supported model IDs.

Usage

Run the MCP server:

python server.py

This will start the server using stdio transport.

Invoking the deepthinking tool

From an MCP-compatible client, you can invoke the deepthinking tool with a query string:

response = mcp.call_tool("deepthinking", {"query": "Explain the theory of relativity step by step."})
print(response)

The tool will return a detailed reasoning response generated by the Groq LLM.

Tool Details

deepthinking(query: str) -> str

A tool that helps AI perform deep thinking (reasoning) processes. It can be used for complex problem solving, planning, multi-step reasoning, and more.

• Arguments:
  • query (str): The input prompt or question.
• Returns:
  • str: The generated reasoning response, ending with </think>.

License

MIT License

Contact

For questions or support, please contact the maintainer.