Ragstack MCP

What HatmanStack/ragstack-mcp MCP server does

HatmanStack/ragstack-mcp MCP server exposes an existing RAGStack knowledge base to MCP-compatible AI assistants. RAGStack is an AWS-based application for ingesting documents, images, video, audio, and web content, then making that material searchable. The server is intended for assistants that need to retrieve information from the deployed knowledge base and use that content in conversations.

Supported content includes office documents, PDFs, HTML, CSV, JSON, XML, EML, EPUB, images, Markdown, video, and audio. Image files can be indexed with visual content and captions. Video and audio are transcribed with AWS Transcribe and divided into 30-second searchable segments, with timestamp information available for media sources.

How it works

The MCP process sends requests to the RAGStack GraphQL endpoint. Authentication uses the x-api-key value generated in the RAGStack dashboard. The deployed RAGStack application handles storage, processing, extraction, embeddings, retrieval, and answer generation. Amazon Bedrock can provide retrieval and AI-generated responses, while S3 and DynamoDB support the underlying serverless pipeline.

After ingestion, files move through upload, processing, and indexing stages. Text commonly takes one to five minutes to index, OCR workloads typically take two to fifteen minutes, and media processing can take five to twenty minutes. Search results can use document metadata filters, and chat responses can include source attribution. Media citations may include links to specific timestamps.

Setup and configuration

Install the package with Python’s package tooling, or run it through uvx without a separate installation:

pip install ragstack-mcp

An MCP client configuration needs the server command and two environment variables:

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

The endpoint and key must come from a deployed RAGStack instance. The main RAGStack project requires an AWS account and documents deployment through AWS Marketplace, CloudFormation, or source-based deployment. Its source deployment prerequisites include Python 3.13 or later, Node.js 24 or later, AWS CLI, SAM CLI, Docker, and uv.

Tools and capabilities

HatmanStack/ragstack-mcp MCP server supports workflows built around the RAGStack knowledge base, including:

• Searching indexed knowledge-base content.
• Asking questions that produce AI-generated answers grounded in retrieved content.
• Uploading documents and media for processing.
• Scraping websites into the knowledge base.
• Analyzing and filtering document metadata.
• Working with OCR-extracted text from supported image and PDF formats.
• Searching transcribed audio and video with timestamp-aware sources.

The wider RAGStack application also supports document reprocessing, reindexing, and deletion, although the supplied material does not define the exact MCP tool names or expose which of these operations are available through this package.

Limitations and notes

This package is an interface to a deployed RAGStack backend, not a standalone local vector index. The backend requires AWS services and configuration, and processing is asynchronous. API access uses a server-side API key; the web component uses IAM authentication instead, which is a separate integration path.

The project is marked as being in active development, so deployment and configuration details may change. Supported client examples include Claude Desktop, Cursor, VS Code, and Amazon Q CLI, along with other MCP-compatible tools. The repository is licensed under Apache 2.0.