Search Engine with RAG and MCP

What is Search Engine with RAG and MCP?

Search Engine with RAG and MCP is a search engine that combines LangChain, Model Context Protocol (MCP), Retrieval-Augmented Generation (RAG), and Ollama to create an agentic AI system capable of searching the web, retrieving information, and providing relevant answers.

How to use Search Engine with RAG and MCP?

Install dependencies with pip or Poetry, then run the application in one of three modes: direct search (python -m src.core.main "query"), agent mode (--agent), or MCP server mode (--server). Optionally configure host and port for server mode, and set up Ollama for local LLM usage.

Key features of Search Engine with RAG and MCP

• Web search via Exa API and content retrieval via FireCrawl
• RAG (Retrieval-Augmented Generation) for relevant information extraction
• MCP server for standardized tool invocation
• Support for local LLMs (Ollama) and cloud LLMs (OpenAI)
• Three operation modes: direct search, agent, or server
• Asynchronous processing for efficient web operations

Use cases of Search Engine with RAG and MCP

• Perform direct web searches and retrieve summarized answers
• Deploy an agentic AI that uses search and RAG tools independently
• Run an MCP server that exposes search and retrieval tools to MCP clients
• Combine local LLMs with external web data for privacy-sensitive queries

FAQ from Search Engine with RAG and MCP

What runtime and dependencies are required?

Python 3.13+ is required. Dependencies include LangChain, MCP libraries, embeddings, FAISS, and API clients for Exa and FireCrawl. Optional: Ollama for local LLM, OpenAI for cloud LLM.

Where does data (embeddings, logs) live?

The project creates data/ directories for data storage and logs/ for log files (auto-created). FAISS vector stores and document chunks are managed locally.

What API keys are needed?

API keys for Exa and FireCrawl are mandatory. An OpenAI API key or Ollama local installation is optional depending on LLM choice.

How do I switch between local and cloud LLMs?

Set the appropriate environment variables in .env. For Ollama, set OLLAMA_BASE_URL and OLLAMA_MODEL; for OpenAI, set OPENAI_API_KEY.

What transports and auth does the MCP server use?

The MCP server listens on a configurable host and port (default: likely localhost:8000). No authentication mechanism is mentioned in the README.