A framework for conversational AI assistance in BIM-based MEP coordination: an MCP-Navisworks implementation

Purpose This paper aims to propose and validate a framework for conversational AI assistance in BIM-based MEP coordination through a proof-of-concept prototype integrating MCP with Autodesk Navisworks, addressing the post-detection decision support gap. Design/methodology/approach A Design Science Research approach guided development and validation of the framework as the primary artifact, instantiated as an MCP-Navisworks prototype and evaluated through three tests on a real residential building project to identify conditions of use rather than generalizable performance. Findings The framework demonstrated technical viability across all tests. Intelligent grouping produced 20 prioritized element groups from 45 unique clashes, though four clashes were assigned to two groups simultaneously. Across the three resolution cases, the assistant surfaced the coordinator’s adopted solution among its suggested alternatives, though alongside non-adopted options and, for the most complex clash, only after explicit dimensional data was supplied. Research limitations/implications Findings establish technical viability rather than generalizable performance. Performance is constrained by input data quality and current LLM spatial reasoning capabilities. No timing or efficiency measurements were collected. Practical implications Natural-language interaction offers a potential route to supporting post-detection clash analysis. Originality/value Building on an open-source MCP-Navisworks base limited to basic model communication, the study’s specific technical contribution is a Clash Detective integration enabling a conversational AI to read and analyze native clash results, through a three-layer architecture validated as an exploratory proof of concept documenting where LLM-based assistance helps and where it fails.