Local RAG Web Search MCP Server

Local RAG Web Search provides a lightweight, privacy-focused solution for retrieving up-to-date information from the web without relying on external APIs. It uses DuckDuckGo to search the web, extracts relevant content from search results, and ranks them using Google's MediaPipe Text Embedder to provide the most relevant context to your LLM. This tool enables your AI assistant to access current information beyond its training cutoff, enhancing responses with fresh knowledge while maintaining privacy by running entirely on your local machine. The implementation is intentionally "primitive" but effective, making it accessible and easy to deploy.

Overview

Local RAG Web Search is a Model Context Protocol (MCP) server that enables your AI assistant to search the web for up-to-date information. It implements a Retrieval-Augmented Generation (RAG) approach that:

Searches DuckDuckGo for your query
Fetches the top 10 search results
Computes embeddings using Google's MediaPipe Text Embedder
Ranks entries based on similarity to your query
Extracts relevant context from the top results
Returns the information to your AI assistant in markdown format

All of this happens locally on your machine without requiring any external API keys.

Installation Options

You can install Local RAG Web Search using one of the following methods:

Option 1: Using Docker (Recommended)

This is the simplest approach if you have Docker installed:

Make sure you have Docker installed on your system
Add the following configuration to your MCP server configuration file:

{
  "mcpServers": {
    "mcp-local-rag": {
      "command": "docker",
      "args": [
        "run",
        "--rm",
        "-i",
        "--init",
        "-e",
        "DOCKER_CONTAINER=true",
        "ghcr.io/nkapila6/mcp-local-rag:latest"
      ]
    }
  }
}

Option 2: Using Python with uv

This approach requires uv to be installed on your system.

Direct Installation via uvx

Add the following to your MCP configuration:

{
  "mcpServers": {
    "mcp-local-rag":{
      "command": "uvx",
      "args": [
        "--python=3.10",
        "--from",
        "git+https://github.com/nkapila6/mcp-local-rag",
        "mcp-local-rag"
      ]
    }
  }
}

Manual Installation

Clone the repository:

git clone https://github.com/nkapila6/mcp-local-rag

Add the following to your MCP configuration:

{
  "mcpServers": {
    "mcp-local-rag": {
      "command": "uv",
      "args": [
        "--directory",
        "/path/to/mcp-local-rag/",
        "run",
        "src/mcp_local_rag/main.py"
      ]
    }
  }
}

Replace /path/to/mcp-local-rag/ with the actual path where you cloned the repository.

Using Local RAG Web Search

Once installed, your AI assistant will automatically use Local RAG Web Search when it needs to retrieve current information from the web.

When you ask a question that requires up-to-date information (beyond the model's training data), the assistant will recognize the need to search the web and will use this tool to fetch relevant information.

You don't need to explicitly invoke the tool - the AI assistant will determine when web search is necessary and will use the tool automatically.

Finding Your MCP Configuration File

The location of your MCP configuration file depends on your operating system:

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

For more information about MCP configuration, visit https://modelcontextprotocol.io/quickstart/user.

Local RAG Web Search MCP Server

web_search