n8n MCP Server with Retrieval Augmented Generation (RAG)

This n8n workflow sets up an MCP (Model Context Protocol) server that can answer questions using Retrieval Augmented Generation (RAG). It leverages Ollama for embeddings and Qdrant as a vector store to provide contextually relevant answers based on submitted documents.

What it does

This workflow automates the process of ingesting documents into a vector store and then using that data to answer questions via an MCP server:

1. Listens for Document Submissions: It provides a form endpoint (On form submission) where users can upload documents.
2. Loads and Processes Documents: When a document is submitted, the Default Data Loader node processes the document content.
3. Generates Embeddings: The Embeddings Ollama node creates vector embeddings for the processed document content. This requires an Ollama server running with a compatible embedding model.
4. Stores Embeddings in Qdrant: The generated embeddings are then stored in a Qdrant Vector Store for efficient similarity search.
5. Exposes an MCP Server: The MCP Server Trigger node sets up an endpoint that listens for questions.
6. Answers Questions with RAG: When a question is received by the MCP server, it will perform a similarity search in the Qdrant vector store using the question's embedding (generated by Ollama) to retrieve relevant document chunks, and then use these chunks as context to generate an answer.

Prerequisites/Requirements

• n8n Instance: A running n8n instance.
• Ollama Server: An accessible Ollama server running with an embedding model (e.g., nomic-embed-text) and a large language model (LLM) for answering questions (e.g., llama2).
• Qdrant Instance: An accessible Qdrant vector database instance.
• Credentials:
  • Ollama Credentials: For the Embeddings Ollama node to connect to your Ollama server.
  • Qdrant Credentials: For the Qdrant Vector Store node to connect to your Qdrant instance.

Setup/Usage

1. Import the Workflow:

   • Download the provided JSON file.
   • In your n8n instance, go to "Workflows" and click "New".
   • Click the three dots menu (...) and select "Import from JSON".
   • Paste the JSON content or upload the file.

2. Configure Credentials:

   • Open the Embeddings Ollama node and configure your Ollama credentials (e.g., server URL, model name).
   • Open the Qdrant Vector Store node and configure your Qdrant credentials (e.g., host, API key, collection name).

3. Activate the Workflow:

   • Toggle the workflow to "Active" in the top right corner of the n8n editor.

4. Ingest Documents:

   • The On form submission node provides a URL. Access this URL in your browser or via an HTTP POST request to submit documents (e.g., text, PDF, Markdown files, depending on the Default Data Loader configuration).
   • Each submitted document will be processed, embedded, and stored in Qdrant.

5. Query the MCP Server:

   • The MCP Server Trigger node will provide an endpoint URL.
   • You can send questions to this endpoint (e.g., via a custom application, another n8n workflow, or curl) to get answers augmented by the documents you've ingested.

This setup allows you to build a powerful question-answering system that can draw information from your own custom knowledge base.