Serverless MCP server for AI agent data lake access on Amazon ECS Fargate

This article describes a serverless Model Context Protocol (MCP) server built on Amazon ECS Fargate that acts as a centralized gateway for AI agents to access enterprise data lakes.

• Configuration-driven data lake access: adding datasets requires only a JSON metadata file, not custom code
• Metadata catalog stores schema descriptions, column semantics, and table relationships for AI agents to reference during SQL generation
• Four-layer architecture: client layer (authentication/session), MCP server layer (orchestration), tools layer (SQL/search), and knowledge-base refresh layer
• Centralized security with Amazon Cognito authentication, AWS WAF, Amazon Bedrock Guardrails, and SELECT-only query enforcement
• Multi-tenant support: any MCP-compatible client can invoke the same tools through standardized /mcp endpoint
• Serverless scaling: ECS Fargate auto-scales client and MCP layers; Athena and Bedrock scale automatically
• End-to-end query latency: 15–20 seconds from user question to natural language response

The metadata-driven approach eliminates per-dataset integration code and enables rapid onboarding of new data sources while maintaining centralized access control and audit logging.