Agentic AI for Autonomous Construction Issue Detection

Quality management remains a persistent challenge in the construction industry, with rework accounting for significant cost growth and schedule delays. While digital inspection technologies have improved data capture, defect identification still relies on manual, error-prone processes. This paper introduces an agentic AI framework, implemented through Model Context Protocol (MCP), for autonomous construction quality management. The framework integrates retrieval-augmented reasoning over specifications and drawings, multimodal site imagery, and structured issue generation through orchestrated large language model (LLM) agents. A case study on an office renovation project demonstrates the feasibility of the approach. The system accurately retrieved construction information from specifications, achieving a mean cosine similarity of 0.82 on queries compared to manual responses and successfully identified 9 of 10 documented quality issues, including design misalignments and field non-conformances. These findings provide empirical evidence that agentic AI can extend beyond descriptive reporting toward actionable quality management. The contributions are twofold: (i) a modular MCP-based architecture for orchestrating construction AI agents, and (ii) validation through deployment in a live project environment. While evaluation is limited in scope, results highlight both the promise and challenges of integrating agentic AI into daily workflows, paving the way for larger-scale studies and future digital twin integration.