RAG Search MCP Server

A Model Context Protocol (MCP) server that provides RAG (Retrieval-Augmented Generation) search capabilities for the "Attention Is All You Need" paper using LlamaIndex and OpenAI.

Overview

This server creates a searchable knowledge base from the seminal Transformer paper "Attention Is All You Need" and exposes it through an MCP tool. It allows you to query the paper's content about transformer architecture and attention mechanisms.

Features

• RAG Search: Query the "Attention Is All You Need" paper using natural language
• Vector Store Index: Uses LlamaIndex to create efficient vector embeddings
• MCP Integration: Seamlessly integrates with MCP-compatible clients
• OpenAI Integration: Leverages OpenAI's models for search and retrieval

Prerequisites

• Python 3.8+
• OpenAI API key
• The "Attention Is All You Need" PDF file

Installation

1. Clone this repository:

git clone <your-repo-url>
cd rag-search-mcp-server

2. Install dependencies:

pip install fastmcp python-dotenv llama-index

3. Create a .env file in the root directory:

OPENAI_API_KEY=your_openai_api_key_here

Important: The .env file should be placed in the same directory as your custom_mcp_server_rag_tool.py file. This file stores your sensitive API credentials and should never be committed to version control. Add .env to your .gitignore file.

4. Download the "Attention Is All You Need" paper and place it at:

C:\Users\spacecat\Downloads\attention_is_all_you_need.pdf

Or update the path in custom_mcp_server_rag_tool.py to match your file location.

Usage

Starting the Server

Run the server:

python custom_mcp_server_rag_tool.py

The server will:

1. Load the PDF document
2. Create a vector store index from the document content
3. Start the MCP server with the RAG search tool available

Connecting to Claude Desktop

To use this MCP server with Claude Desktop:

1. Configure Claude Desktop: Add the server to your Claude Desktop configuration file:

   Windows: %APPDATA%\Claude\claude_desktop_config.json macOS: ~/Library/Application Support/Claude/claude_desktop_config.json Linux: ~/.config/Claude/claude_desktop_config.json

2. Add server configuration:

   {
     "mcpServers": {
       "rag-search": {
         "command": "python",
         "args": ["path/to/your/custom_mcp_server_rag_tool.py"],
         "env": {
           "OPENAI_API_KEY": "your_openai_api_key_here"
         }
       }
     }
   }
   

   Alternative (using .env file):

   {
     "mcpServers": {
       "rag-search": {
         "command": "python",
         "args": ["path/to/your/custom_mcp_server_rag_tool.py"]
       }
     }
   }
   

   (Make sure your .env file is in the same directory as the Python script)

3. Restart Claude Desktop for the changes to take effect.

4. Verify connection: You should see the RAG search tool available in Claude Desktop's interface.

Using the RAG Search Tool

Once connected to an MCP client, you can use the search_rag tool to query the paper:

Example queries:

• "What is attention mechanism?"
• "How does multi-head attention work?"
• "Explain the transformer architecture"
• "What are the advantages of self-attention?"

Code Structure

custom_mcp_server_rag_tool.py
├── Environment Setup (OpenAI API key)
├── AppContext (Data class for storing the index)
├── Lifespan Manager (Loads PDF and creates index)
├── MCP Server Setup
└── RAG Search Tool (Query interface)

Key Components

• FastMCP: The MCP server framework
• SimpleDirectoryReader: Loads PDF documents
• VectorStoreIndex: Creates searchable vector embeddings
• Query Engine: Processes natural language queries

Configuration

Environment Variables

Variable	Description	Required
OPENAI_API_KEY	Your OpenAI API key for embeddings and LLM	Yes

File Paths

Update the PDF file path in the lifespan function:

documents = SimpleDirectoryReader(input_files=["path/to/your/pdf"]).load_data()

API Reference

Tools

search_rag

• Description: Search content in the PDF about transformer and attention methodology
• Parameters:
  • message (string): The search query
• Returns: String response with relevant information from the paper

Dependencies

• fastmcp: MCP server framework
• python-dotenv: Environment variable management
• llama-index: RAG and vector search capabilities
• openai: OpenAI API integration (via llama-index)

Troubleshooting

Common Issues

1. Missing OpenAI API Key

   • Ensure your .env file contains a valid OPENAI_API_KEY
   • Check that the key has sufficient credits
   • If using Claude Desktop config, verify the API key is correctly set in the JSON

2. PDF File Not Found

   • Verify the PDF path in the lifespan function
   • Ensure the file exists and is readable

3. Import Errors

   • Install all required dependencies: pip install fastmcp python-dotenv llama-index

4. Claude Desktop Connection Issues

   • Verify the path to custom_mcp_server_rag_tool.py is correct in the config
   • Ensure Python is in your system PATH
   • Check Claude Desktop logs for error messages
   • Restart Claude Desktop after configuration changes

5. Server Not Starting

   • Make sure all dependencies are installed
   • Check that the PDF file path is accessible
   • Verify OpenAI API key is valid and has credits

Contributing

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Submit a pull request

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• Original "Attention Is All You Need" paper by Vaswani et al.
• LlamaIndex for RAG capabilities
• FastMCP for the MCP server framework