home / mcp / ragstack mcp server

RAGStack MCP Server

Provides an MCP server to query RagStack knowledge base via AI assistants with retrieval and source citations.

Installation

Add the following to your MCP client configuration file.

Configuration

{
  "mcpServers": {
    "hatmanstack-ragstack-lambda": {
      "command": "pip",
      "args": [
        "install",
        "ragstack-mcp"
      ],
      "env": {
        "RAGSTACK_API_KEY": "YOUR_API_KEY",
        "RAGSTACK_GRAPHQL_ENDPOINT": "YOUR_ENDPOINT"
      }
    }
  }
}

You can deploy and use a serverless knowledge base and AI chat system that processes documents, images, video, and audio, extracts text with OCR or transcription, and lets you search with context-aware answers and source citations. It’s designed to scale with cloud-native services and supports multimodal embeddings, media transcription, and retrieval-augmented chat.

How to use

You interact with the MCP server through an integration that connects your AI assistant to the knowledge base. Upload documents and media to build the knowledge base, then ask questions. The chat responses include sources and timestamps when you query media, so you can verify the information and jump to the exact moment in a video or audio file.

How to install

Prerequisites: you need an environment with Python and a package manager, plus access to run MCP configurations in your AI assistant ecosystem.

# Technique 1: Install the MCP integration tool locally (example)
pip install ragstack-mcp

# Technique 2: Or configure your MCP with the provided JSON snippet using your assistant's settings

MCP Server integration for AI assistants

To connect your AI assistant to the knowledge base, add the MCP config to your assistant’s MCP setup. This enables querying the knowledge base directly from your assistant with the appropriate authentication.

{
  "ragstack-kb": {
    "command": "uvx",
    "args": ["ragstack-mcp"],
    "env": {
      "RAGSTACK_GRAPHQL_ENDPOINT": "YOUR_ENDPOINT",
      "RAGSTACK_API_KEY": "YOUR_API_KEY"
    }
  }
}

Notes on deployment and usage

The MCP server is designed to operate in a serverless architecture, leveraging Lambda, Step Functions, S3, and DynamoDB for scalable document processing, embedding generation, and retrieval. You can run in demo mode for evaluation or in production with your own endpoints and keys.

Additional configuration and troubleshooting

- Ensure your endpoint and API key are correctly configured in the MCP env vars. - When testing media transcription, verify that the language and speaker diarization settings meet your needs. - Use the knowledge base reindex option if you update metadata or processing settings and need fresh results.

Available tools

ragstack-mcp

MCP server component that enables retrieval-augmented querying of the RagStack knowledge base from AI assistants.