MCP-RAGNAR - a local RAG MCP Server

A local MCP server that implements RAG (Retrieval-Augmented Generation) with sentence window retrieval.

Features

• Document indexing with support for multiple file types (txt, md, pdf, doc, docx)

• Sentence window retrieval for better context understanding

• Configurable embedding models (OpenAI or local hugging face mode - i.e BAAI/bge-large-en-v1.5)

• MCP server integration for easy querying

Related MCP server: MCP-Markdown-RAG

Requirements

• Python 3.10+

• UV package manager

Installation

1. Clone the repository:

git clone <repository-url>
cd mcp-ragnar

2. Install dependencies using UV:

uv pip install -e .

Usage

Indexing Documents

You can index documents either programmatically or via the command line.

Indexing

python -m indexer.index /path/to/documents /path/to/index

# to change the default local embedding model and chunk size
python -m indexer.index /path/to/documents /path/to/index --chunk-size=512 --embed-model BAAI/bge-small-en-v1.5

# With OpenAI embedding endpoint (put your OPENAI_API_KEY in env)
python -m indexer.index /path/to/documents /path/to/index --embed-endpoint https://api.openai.com/v1 --embed-model text-embedding-3-small --tokenizer-model o200k_base

# Get help
python -m indexer.index --help

Running the MCP Server

Configuration

can be supplied as env var or .env file

• EMBED_ENDPOINT: (Optional) Path to an OpenAI compatible embedding endpoint (ends with /v1). If not set, a local Hugging Face model is used by default.

• EMBED_MODEL: (Optional) Name of the embedding model to use. Default value of BAAI/bge-large-en-v1.5.

• INDEX_ROOT: The root directory for the index, used by the retriever. This is mandatory for MCP (Multi-Cloud Platform) querying.

• MCP_DESCRIPTION: The exposed name and description for the MCP server, used for MCP querying only. This is mandatory for MCP querying. For example: "RAG to my local personal documents"

• INDEX_ROOT: the root path of the index

in SSE mode it will listen to http://localhost:8001/ragnar

python server/sse.py

in stdio mode

install locally as an uv tool

uv tool install .

Claude Desktop:

Update the following:

On MacOS: ~/Library/Application\ Support/Claude/claude_desktop_config.json

On Windows: %APPDATA%/Claude/claude_desktop_config.json

Example :

{
  "mcpServers": {
    "mcp-ragnar": {
      "command": "uvx",
      "args": [
        "mcp-ragnar"
      ],
      "env": {
        "OPENAI_API_KEY": "",
        "EMBED_ENDPOINT": "https://api.openai.com/v1",
        "EMBED_MODEL": "text-embedding-3-small",
        "MCP_DESCRIPTION": "My local Rust documentation",
        "INDEX_ROOT": "/tmp/index"
      }
    }
  }
}

License

GNU General Public License v3.0