Beyond ETL: A Lean AI Data Warehouse Architecture Using Model Context Protocol with Integrated Data Governance

Abstract

Modern enterprises face increasing challenges in managing data infrastructure for AI-driven analytics while maintaining governance, compliance, and cost efficiency. This paper proposes a Lean AI Data Warehouse (LAIDW) architecture that integrates the Model Context Protocol (MCP) as a unifying interface layer between AI agents and heterogeneous data sources. The proposed framework reduces redundant data movement by enabling AI models to query data sources directly through standardized MCP connectors, reducing intermediate storage overhead and improving data freshness. A complementary Data Governance Framework (DGF) enforces data quality, lineage tracking, access control, and auditability across the MCP-connected ecosystem. Comparative analysis against conventional ETL-based architectures indicates potential structural advantages in storage efficiency (estimated 60-70% reduction), query freshness (near real-time versus batch-delayed), and governance coverage (automated versus manual lineage capture), while recognizing that end-to-end latency and throughput depend on source-system performance, network conditions, query complexity, and selective materialization for complex multi-source analytical workloads. The proposed LAIDW-MCP-DGF architecture offers a scalable, maintainable approach for organizations seeking to operationalize AI workflows with minimal data infrastructure overhead.