MCP RAG

MCP RAG is an MCP server that provides managed Retrieval-Augmented Generation capabilities by connecting MCP clients directly to Amazon Bedrock Knowledge Bases and Amazon OpenSearch. Designed for cloud engineers, AI developers, and system architects, it streamlines document processing pipelines and semantic vector retrieval across AWS infrastructure. When users upload files to Amazon S3, an automated event pipeline processes pages, executes contextual embeddings, extracts text and summaries from images using multimodal models, and indexes data into OpenSearch or Bedrock Knowledge Bases. Through Model Context Protocol tools, client agents can execute semantic searches, retrieve contextual chunks, and integrate parent-child chunking architectures. MCP RAG offers multiple integration modes, including native AWS Bedrock Knowledge Base tooling, custom Python MCP servers, and AWS Lambda-based search handlers. This allows developers to supply LLMs with rich, situated context from complex multi-page documents, tables, and images without managing ad-hoc vector retrieval logic.

Category: AI Memory & Context

Tags: embeddings, knowledge-base, opensearch, rag, retrieval

Visit MCP RAG

How to install and configure MCP RAG

1. Clone the repository from https://github.com/kyopark2014/mcp-rag and install necessary dependencies using Python. 2. Configure your AWS credentials with sufficient permissions to access Amazon Bedrock and OpenSearch. 3. To use the custom knowledge base tool, configure your MCP client configuration file (such as Claude Desktop) under the mcpServers section: json { "mcpServers": { "knowledge_base_custom": { "command": "python", "args": [ "application/mcp_server_knowledge_base.py" ], "env": { "KB_INCLUSION_TAG_KEY": "mcp-rag" } } } } 4. Alternatively, deploy the AWS Bedrock retrieval server using uvx awslabs.bedrock-kb-retrieval-mcp-server@latest as described in the repository. 5. Restart your MCP client to enable the RAG tools.

What you can do with MCP RAG

• Querying Amazon Bedrock Knowledge Bases directly from an MCP client using semantic search keywords. * Retrieving detailed parent chunks from OpenSearch based on child chunk matches during contextual question answering. * Ingesting multi-page document images into OpenSearch with automated multimodal OCR and contextual summaries via S3 events. * Executing custom search workflows through AWS Lambda functions exposed as Model Context Protocol tools. * Integrating advanced RAG pipelines with LangGraph or EC2-hosted AI applications using standardized MCP interfaces.

Key facts

• https://github.com/kyopark2014/mcp-rag
• AI Memory & Context, Databases & Data Stores
• embeddings, knowledge-base, opensearch, rag, retrieval

Part of MCP Servers

What is MCP RAG?

MCP RAG is an open-source reference architecture and MCP server setup that connects AI assistants to Amazon Bedrock Knowledge Bases and Amazon OpenSearch. It provides advanced retrieval tools, automated document ingestion pipelines, contextual chunking, and multimodal OCR capabilities.

Which MCP clients work with MCP RAG?

MCP RAG works with any client supporting the Model Context Protocol, including Claude Desktop, LangGraph workflows, and custom AI applications running on Amazon EC2 or local development environments.

How does MCP RAG handle complex documents?

It processes document pages as images, generating multimodal summaries and contextual text before saving them into OpenSearch. Additionally, it applies parent-child chunking so search matches on small chunks return full parent context.

What AWS services are required for MCP RAG?

The architecture relies on Amazon S3 for document ingestion, AWS Lambda and Amazon SQS for asynchronous event processing, and either Amazon Bedrock Knowledge Bases or Amazon OpenSearch Service for vector storage.