Request for Feedback: Centralized Knowledge Base Access via MCP
Hi everyone, We’re exploring a potential architecture to streamline how multiple organizations access a shared knowledge base (KB) for GenAI-based applications—particularly in regulated or collaborative industry environments. Today, we’re seeing several organizations independently build GenAI solutions based on Retrieval-Augmented Generation (RAG), each uploading and maintaining very similar document sets. This results in duplicated effort, inconsistent knowledge curation, and fragmented governance. We're considering a more efficient and scalable alternative: Centralize shared documents into a single Amazon Bedrock Knowledge Base in one AWS account, and expose access to that KB via a REST API using the Amazon Bedrock Knowledge Base Retrieval MCP Server. In this model:
- Each organization can maintain its GenAI application in its own cloud environment (AWS or otherwise).
- Applications would query the centralized KB via REST through the MCP Server, eliminating the need for cross-account IAM roles, STS configurations, and tight coupling to AWS infrastructure.
- Each application could still use local tools or a complementary private KB when needed, combining them through agents that dynamically choose the appropriate tool for each query.
This approach allows for:
- Simplified integration for multi-cloud and multi-org environments
- Standardized, secured, and auditable access to shared knowledge
- Lower operational overhead compared to cross-account architectures
From your experience, do you believe this architecture using the MCP Server would be feasible by leveraging MCP’s own resource discovery and retrieval capabilities to expose, navigate, and query the centralized knowledge base?
- Language
- English
asked a year ago300 views
- Newest
- Most votes
- Most comments
Based on the information available, I don't see any references to an "MCP Server" or "Amazon Bedrock Knowledge Base Retrieval MCP Server" in the context of Amazon Bedrock Knowledge Bases.
However, I can address the core concept of your proposed architecture - centralizing a shared knowledge base with Amazon Bedrock.
Amazon Bedrock Knowledge Bases do provide a fully managed capability that supports the entire RAG workflow, from data ingestion to retrieval and prompt augmentation. The service allows you to connect to various data sources, process documents, create embeddings, and store them in vector databases like Amazon OpenSearch Serverless, Pinecone, or Redis Enterprise Cloud.
For your proposed centralized architecture, Amazon Bedrock Knowledge Bases offer several relevant features:
1. API-based access to the knowledge base, which could potentially be exposed through API Gateway to create the REST API interface you're envisioning
2. Enterprise-grade security features including encryption and access controls to protect sensitive data
3. Seamless integration with other Amazon Bedrock services like models and agents
4. Session context management for multi-turn conversations
5. Citation tracking for transparency and credibility
Your approach of centralizing shared documents into a single Amazon Bedrock Knowledge Base and exposing it via a REST API could indeed provide the benefits you mentioned - standardized access to shared knowledge, reduced duplication of effort, and consistent governance. Organizations could maintain their GenAI applications in their own environments while accessing the centralized knowledge base.
However, you would need to design the API access layer yourself, as there isn't a specific "MCP Server" component mentioned in the Amazon Bedrock documentation. You might consider using Amazon API Gateway with AWS Lambda to implement the fully managed backend API endpoint, which is a pattern mentioned in AWS Bedrock Generative AI application architectures.
Sources
Community | Unleashing the Power of AWS Bedrock-Knowledge Base for Generative AI - Claude2 Model
Knowledge bases for Amazon Bedrock - AWS Prescriptive Guidance
Community | Choose the right RAG Architecture for your AWS GenAI App
Community | AWS Bedrock Generative AI Application Architecture
How Amazon Bedrock knowledge bases work - Amazon Bedrock
answered a year ago
Relevant content
asked 2 months ago
