MCP Local Rag

🚀 mcp-local-rag

A locally run "raw" RAG-like web search model context protocol (MCP) server. ✨ No API required! ✨

🚀 Quick Start

This project provides a locally run MCP server for web search, eliminating the need for an API.

📦 Installation

1. You need to install uv: https://docs.astral.sh/uv/

Option 1: Without Cloning in Step 2

Just paste the following directly into the Claude configuration. You can find the configuration path here: https://modelcontextprotocol.io/quickstart/user

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

Option 2: Cloning the Repository

2. Clone this GitHub repository (optional, you can skip the above configuration).

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

3. Add the following to the Claude configuration. You can find the configuration path here: https://modelcontextprotocol.io/quickstart/user

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

💻 Usage Examples

Basic Usage

When asked to fetch/search the web, the model will prompt you to use the MCP server for chatting.

In the example, a question was asked about Google's latest Gemma model released yesterday, which is new information that Claude is not aware of.

Result

The results from the local rag_search help the model answer with new information.