Conceptualising RAG-Driven Agentic AI with Multi-Layer MCP for Seismic Structural Systems

The integration of Generative AI into civil engineering is currently constrained by the susceptibility of Large Language Models (LLMs) to hallucination and their inherent lack of physics-based knowledge. To address these limitations, this paper presents a conceptual framework for the integration of Agentic Artificial Intelligence (AI) into the complete lifecycle of seismic-resistant structural engineering. The proposal employs a modular software architecture built on the Model Context Protocol (MCP), enabling distributed collaboration among specialised AI agents across six critical stages: (1) seismic hazard assessment, (2) structural modelling and analysis, (3) design and optimisation, (4) construction quality control, (5) structural health monitoring (SHM), and (6) ethical audit and explainability. In this architecture, agents operate as autonomous MCP Clients within a standardised context, orchestrating workflows by communicating directly with deterministic MCP Servers and the human user. This structure strictly manages tool execution through synchronous, verifiable MCP calls, ensuring that stochastic agentic reasoning remains decoupled from immutable numerical execution. By grounding generative outputs in physics-based engines and Retrieval-Augmented Generation (RAG), the framework ensures traceable reasoning, transparency, and professional accountability, offering a pathway for the ethical deployment of AI systems in civil and structural engineering.