Watsonx Rag MCP Server

🚀 Complete Guide to Building a RAG Server with Watsonx.ai and Integrating it into Claude Desktop via MCP Protocol

This project aims to demonstrate how to leverage Watsonx.ai to build a fully - functional Retrieval Augmented Generation (RAG) server and achieve persistent vector storage through ChromaDB. Meanwhile, we'll also show how to integrate this server into Claude Desktop via the Model Context Protocol (MCP), enabling it to answer questions based on the content of PDF documents.

✨ Features

1. Build a RAG Server: Use Watsonx.ai and ChromaDB to create a RAG server for parsing and understanding PDF files and generating answers based on their content.
2. Configure MCP Protocol Integration: Register a custom Python tool (our RAG server) as an available tool by editing Claude Desktop's configuration file.
3. Verification and Usage: After ensuring the correct configuration, users can call this tool in Claude Desktop to ask and answer questions based on PDF content.

🚀 Quick Start

Step 1: Build the RAG Server

1. Install Necessary Libraries

pip install watsonx chromadb langchain

2. Initialize the Watsonx Model

Initialize the model in the server.py file:

from langchain.embeddings import Embeddings
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.vectorstores import ChromaDB
from langchain.document_loaders import PDFLoader
import os

# Initialize Watsonx embeddings
embeddings = Embeddings()

3. Load and Process Documents

# Load the PDF file
loader = PDFLoader("manual.pdf")
documents = loader.load()

# Split the document into small chunks
text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
texts = text_splitter.split_documents(documents)

# Create a ChromaDB storage
vectorstore = ChromaDB.from_texts(
    texts,
    embeddings,
    persist_directory="chroma_store",
    collection_name="manual_vectors"
)

4. Define the Query Function

def chat_with_manual(query: str) -> str:
    docs_and_scores = vectorstore.similarity_search_with_score(query, k=3)
    context = "\n".join([doc.page_content for doc, _ in docs_and_scores])
    return f"According to the manual content: {context}"

Step 2: Configure Claude Desktop Integration

1. Find the Location of Claude Desktop's Configuration File

It is usually located in the .config/claude/settings folder under the user's home directory.

2. Edit the Configuration File

Add the following content to the claude_desktop_config.json file:

{
    "mcpServers": {
        " rag_manual_server": {
            "type": "python",
            "command": "python server.py chat_with_manual {query}",
            "description": "Answer questions based on the user manual"
        }
    }
}

Step 3: Verification and Usage

1. Start Claude Desktop and Load the Configuration

Make sure to restart Claude Desktop to apply the new configuration.

2. Test the Tool Call

Enter the following content in the chat window to test the new tool:

/tool rag_manual_server

Then ask a question, for example:

What is the maximum flight time mentioned in the /manual?

The system will generate corresponding answers based on the content of the PDF.

📚 Documentation

⚠️ Important Note

• Environment Configuration: Ensure that all dependencies are correctly installed and the paths are set correctly.
• Performance Optimization: Consider optimizing the efficiency of text splitting and vector storage when dealing with a large number of documents.
• Error Handling: Add exception capture and logging functions in practical applications.

📄 Conclusion

Through this project, we have successfully built a RAG server based on Watsonx.ai and ChromaDB and integrated it into Claude Desktop. This implementation demonstrates how to seamlessly combine advanced NLP technologies with existing tools to provide users with powerful Q&A capabilities.