This AI agent-enabled Smart Water Anomaly Detection Platform addresses the core pain points in smart water management—delayed anomaly response, low data accessibility for non-professionals, and data silos—by building a sensor-based "virtual council" that enables real-time collaborative analysis of urban hydrological environments. It adopts a locally deployed Qwen 14B model (via Ollama) and the MCP protocol to ensure secure data processing, and an optimized MTAD-GAT algorithm featuring a dual-graph structure and dual-head output, which accurately differentiates between sensor technical faults and environmental anomalies through cross-validation among adjacent sensors. Equipped with a user-friendly, multi-device compatible web interface, the platform features an open sensor agent protocol that supports expansion to multi-domain urban monitoring scenarios and cross-agent collaboration. It strives to construct a people-centric smart water system that is perceivable and participatory for all citizens, laying a solid foundation for the intelligent operation of future smart cities.