mcpRAG

What is mcpRAG?

mcpRAG is a Retrieval-Augmented Generation (RAG) server that processes text documents using open‑source embeddings from Ollama (nomic), a FAISS vector database, and the Gemini 2.0‑flash LLM. It is designed for users who want a local RAG pipeline with a cloud‑based LLM.

How to use mcpRAG?

The README does not provide installation, configuration, or invocation instructions. No commands or config keys are mentioned.

Key features of mcpRAG

• Uses nomic embeddings via Ollama locally.
• Uses Gemini 2.0‑flash as the LLM.
• Stores chunks as JSON with file name, chunk ID, and text.
• Indexes embeddings with FAISS and stores locally.
• Supports appending additional text to the existing index.

Use cases of mcpRAG

• Answer questions from a collection of local text documents.
• Incrementally expand a knowledge base by adding new documents.
• Run a RAG pipeline with a fully open‑source embedding and vector store.

FAQ from mcpRAG

What embeddings and LLM does mcpRAG use?

It uses nomic embeddings (run locally with Ollama) and the Gemini 2.0‑flash LLM.

How are documents stored and retrieved?

Text files are chunked into JSON records (file name, chunk ID, text). Each chunk is embedded, indexed by FAISS, and saved locally. Queries produce embeddings that are searched in the FAISS index; the retrieved chunk text is passed to the LLM with the query.

Can I add more documents after the index is built?

Yes. The README states additional text can be appended to the existing index by loading the stored index and embeddings file, then running queries on the updated index.

What vector database is used?

FAISS, an open‑source library for efficient similarity search.

Does mcpRAG require cloud services?

Yes, it requires the Gemini API for the LLM. The embedding step runs locally via Ollama.