Build a semantic layer for agentic AI on AWS with Strategy Mosaic and Amazon Bedrock AgentCore
Strategy Mosaic exposes a unified semantic layer to AI agents via MCP, keeping business logic out of prompts
“The joins, metric definitions, and per-caller row and column filters live inside Mosaic, not in the agent prompt.”
AWS and Strategy detail how to wire a dimensional semantic layer (Mosaic) to a Bedrock AgentCore-hosted Strands agent via the Model Context Protocol, so metric definitions, joins, and row/column security rules stay in the governed data model rather than agent prompts. It is the third installment in a series comparing Stardog (ontology/SPARQL), Denodo (data virtualization), and now Mosaic (dimensional/metric) as semantic-layer options for enterprise agentic AI. The practical insight — that MCP lets agents query a governed business model without embedding SQL or business logic in the prompt — is a useful architectural pattern but the post is a vendor walkthrough rather than a novel industry announcement.