From silos to insights: Federated data access patterns for AI agents

This article presents federated data access patterns using Model Context Protocol (MCP) and Amazon Bedrock AgentCore to enable AI agents to query data across multiple enterprise systems without centralizing data.

• Catalog-first access pattern: MCP servers wrap AWS Glue Data Catalog and Amazon Athena for unified S3 data discovery and querying
• Direct source access pattern: MCP servers query operational databases like Amazon Aurora directly without intermediate catalog layers
• Hybrid access pattern: Combines both catalog-first and direct access patterns under a single orchestrator agent for diverse data landscapes
• Reference architecture spans batch data on S3, streaming data in Kinesis, and relational data in Aurora with unified governance through AWS Lake Formation
• Amazon Bedrock AgentCore Gateway aggregates MCP servers behind a single endpoint for tool discovery, authentication, and routing
• Agents use semantic search and schema discovery workflows to navigate complex data environments without requiring users to know SQL or specific APIs

MCP enables self-service analytics by standardizing how AI agents connect to diverse data sources, eliminating data silos and reducing dependency on data engineers for ad-hoc queries.