atomic-research-mcp

# Atomic Research MCP A powerful web research pipeline MCP built using the [Atomic Agents framework](https://github.com/BrainBlend-AI/atomic-agents). For a full breakdown of the code, please [check out this article](https://medium.com/ai-advances/using-custom-agents-in-cursor-windsurf-copilot-and-others-to-supercharge-your-workflow-f936b630c5e5?sk=30cd72ee79ef8333d5935d556f79471e) For now, this requires an API key for [Tavily](https://tavily.com/) and also [an OpenAI API key ](https://platform.openai.com/api-keys). In the future, I plan to make this more configurable so you can use SearxNG instead of Tavily or use Groq or Anthropic instead of OpenAI, thanks to the Atomic Agents framework this is all easy and possible, it just requires a bit of time and I wanted to get the initial project out there. **Feel free to contribute, though!** ## Support Do you like this project? [Please consider a small donation](https://www.paypal.com/paypalme/KennyVaneetvelde), it means the world to me! ## Overview This project implements an advanced web research pipeline that leverages the Model Context Protocol (MCP) and Atomic Agents to provide comprehensive answers to research questions. The pipeline automates the entire research process: 1. Generating optimized search queries 2. Performing web searches using Tavily 3. Scraping and processing relevant web pages 4. Synthesizing information into coherent answers ## Architecture The system follows a modular architecture based on the MCP client-server model: ### Core Components - **MCP Server**: Implements the Model Context Protocol to expose tools and manage client connections - **Web Search Pipeline**: Orchestrates the research workflow through multiple stages - **Atomic Agents**: Specialized AI agents for query generation and question answering - **Tools**: Reusable components for web search (Tavily) and web page scraping ### Pipeline Flow ``` User Question → Query Generation → Web Search → Content Scraping → Answer Synthesis → Formatted Response ``` ## Technologies - **Model Context Protocol (MCP)**: An open standard for AI applications to access contextual information - **Atomic Agents**: A modular framework for building AI agents with well-defined input/output schemas - **Tavily API**: A specialized search engine for AI applications - **OpenAI**: Powers the underlying language models for query generation and answer synthesis - **Python 3.12+**: The foundation of the application ## Setup ### Prerequisites - Python 3.12 or higher - Tavily API key - OpenAI API key ### Installation 1. Clone the repository: ```bash git clone https://github.com/yourusername/agentic-research-mcp.git cd agentic-research-mcp ``` 2. Install dependencies: ```bash pip install -e . ``` 3. Set up environment variables: ```bash # Create a .env file in the project root echo "TAVILY_API_KEY=your_tavily_api_key" > .env echo "OPENAI_API_KEY=your_openai_api_key" >> .env ``` ## Usage ### Running the Server Start the MCP server: ```bash python -m atomic_research_mcp.server ``` OR ```bash [rootfolder]\atomic-reseearch-mcp\.venv\Scripts\atomic-research ``` OR configure it in, for example, Cursor  ### Testing with the Client Run the test client to verify functionality: ```bash python test_client.py ``` ### Example Output The system returns comprehensive research results including: - Generated search queries - Top search results with relevance scores - A detailed answer synthesized from multiple sources - References to source materials - Suggested follow-up questions ## Project Structure ``` atomic_research_mcp/ ├── agents/ │ ├── query_agent.py # Generates optimized search queries │ └── qa_agent.py # Synthesizes answers from scraped content ├── tools/ │ ├── tavily_search.py # Interface to Tavily search API │ └── webpage_scraper.py # Extracts and processes web content ├── server.py # MCP server implementation └── config.py # Configuration settings ``` ## Configuration The system can be configured through environment variables: - `TAVILY_API_KEY`: Required for web search functionality - `OPENAI_API_KEY`: Required for AI agent operations - `OPENAI_MODEL`: Optional, defaults to "gpt-4o-mini" ## Contributing Contributions are welcome! Please feel free to submit a Pull Request. ## License [MIT License](LICENSE)