From the source
Amazon's North America Sort Center team built a serverless Model Context Protocol server on Amazon ECS Fargate so planning teams can query operational data conversationally.
- The team says the problem was that agents were wired directly to databases and APIs, so each new dataset needed custom code and each agent reimplemented its own access control.
- The server acts as a centralized, agent-agnostic gateway between AI agents and the enterprise data lake, and exposing a new dataset requires only a JSON metadata file that the server discovers on the next pipeline sync.
- The architecture separates client, MCP server, tools and knowledge-base refresh layers, with Amazon Cognito handling authentication, Amazon DynamoDB storing session state, and Amazon Bedrock orchestrating tools built with FastMCP.
- Amazon Athena and Amazon Redshift run SQL against the data lake, while Amazon Bedrock Knowledge Bases handles semantic search and AWS Lambda manages ingestion into Amazon S3.
- A context mapping configuration scopes each query to relevant tables, which the team says kept end-to-end interaction latency in the 15–20 second range in its testing.






