An Ontology-Driven MCP Agent Framework for Ensuring Integrity in Architectural Structural Design

Abstract

The rapid proliferation of Large Language Model (LLM)-based autonomous agents has demonstrated strong potential for automating knowledge-intensive tasks. However, their probabilistic reasoning often leads to hallucinations and weak domain knowledge integrity, limiting their applicability in high-stakes architecture, engineering, and construction (Architecture, Engineering, and Construction (AEC)) contexts where engineering reliability is critical. To address these challenges, this paper proposes a novel architecture, termed the Ontology-driven Model Context Protocol (O-MCP) Framework, which integrates formal ontology with the standardized Model Context Protocol to ensure logical consistency and trustworthy knowledge exchange. The framework introduces a three-layer architecture consisting of a Knowledge and Data Source Layer, a Semantic MCP Server Layer, and an Intelligent Agent Layer. Its core contribution is the Semantic MCP Server, which incorporates a semantic guardrail mechanism that validates agent requests in real time using ontology-based reasoning. A case study in architectural structural design illustrates how the framework formalizes expert tacit knowledge and preserves decentralized knowledge sovereignty. A proof-of-concept experiment in reinforced concrete slab design positions O-MCP as a reference standard for formalized expert knowledge. Results show that while LLM-based RAG systems achieve approximately 83% agreement, systematic errors persist in numerical reasoning and multi-condition rule application, highlighting the value of formal ontological verification for guaranteeing engineering integrity.