RAG-Enabled MCP Server for Semantic Q&A with Qdrant

This n8n workflow creates an MCP Server that leverages a local semantic database (Qdrant) powered by Ollama and embeddings from mxbai-embed-large to perform Retrieval Augmented Generation (RAG). It answers user questions by retrieving relevant information from ingested documents, ensuring contextually accurate responses without relying on cloud services.

The workflow integrates with an MCP Trigger node to receive queries, processes them through semantic search in Qdrant, and generates responses using Llama 3.2. A separate RAG Ingestion Pipeline (linked) allows uploading PDFs to populate the vector database. It's designed for local execution only, prioritizing data privacy and offline capability.

Benefits include enhanced accuracy over basic LLMs, reduced hallucinations via RAG, and seamless integration with other n8n AI agents. Use cases span knowledge bases, internal Q&A systems, document search, and custom AI assistants in domains like research, legal, or customer support.

Setup requires the self-hosted AI starter kit; activate the flow, ingest data, and pair with an MCP Client for querying. Ideal for developers building robust, local AI solutions.