Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MCP4IFC: IFC-Based Building Design Using Large Language Models

Bharathi Kannan Nithyanantham, Tobias Sesterhenn, Ashwin Nedungadi, Sergio Peral Garijo, Janis Zenkner, Christian Bartelt, Stefan Lüdtke

TL;DR

MCP4IFC introduces an open-source MCP server that enables LLMs to directly query, create, and edit IFC-based BIM data through a standardized Model Context Protocol, using IfcOpenShell and Blender (via Bonsai) as the execution environment. The framework combines in-context learning with retrieval-augmented generation to dynamically produce Python code for IFC tasks beyond predefined tools, supporting end-to-end BIM manipulation in a platform-agnostic manner. Through scene querying, semantic editing, and geometry generation experiments, the paper demonstrates feasible end-to-end LLM-driven IFC interactions while highlighting current limitations in semantic robustness and geometric correctness. The work provides a foundation for AI-assisted BIM design, offering an extensible toolkit and a path toward broader tool libraries, generative design capabilities, and multi-modal spatial feedback across IFC-compliant workflows.

Abstract

Bringing generative AI into the architecture, engineering and construction (AEC) field requires systems that can translate natural language instructions into actions on standardized data models. We present MCP4IFC, a comprehensive open-source framework that enables Large Language Models (LLMs) to directly manipulate Industry Foundation Classes (IFC) data through the Model Context Protocol (MCP). The framework provides a set of BIM tools, including scene querying tools for information retrieval, predefined functions for creating and modifying common building elements, and a dynamic code-generation system that combines in-context learning with retrieval-augmented generation (RAG) to handle tasks beyond the predefined toolset. Experiments demonstrate that an LLM using our framework can successfully perform complex tasks, from building a simple house to querying and editing existing IFC data. Our framework is released as open-source to encourage research in LLM-driven BIM design and provide a foundation for AI-assisted modeling workflows. Our code is available at https://show2instruct.github.io/mcp4ifc/.

MCP4IFC: IFC-Based Building Design Using Large Language Models

TL;DR

MCP4IFC introduces an open-source MCP server that enables LLMs to directly query, create, and edit IFC-based BIM data through a standardized Model Context Protocol, using IfcOpenShell and Blender (via Bonsai) as the execution environment. The framework combines in-context learning with retrieval-augmented generation to dynamically produce Python code for IFC tasks beyond predefined tools, supporting end-to-end BIM manipulation in a platform-agnostic manner. Through scene querying, semantic editing, and geometry generation experiments, the paper demonstrates feasible end-to-end LLM-driven IFC interactions while highlighting current limitations in semantic robustness and geometric correctness. The work provides a foundation for AI-assisted BIM design, offering an extensible toolkit and a path toward broader tool libraries, generative design capabilities, and multi-modal spatial feedback across IFC-compliant workflows.

Abstract

Bringing generative AI into the architecture, engineering and construction (AEC) field requires systems that can translate natural language instructions into actions on standardized data models. We present MCP4IFC, a comprehensive open-source framework that enables Large Language Models (LLMs) to directly manipulate Industry Foundation Classes (IFC) data through the Model Context Protocol (MCP). The framework provides a set of BIM tools, including scene querying tools for information retrieval, predefined functions for creating and modifying common building elements, and a dynamic code-generation system that combines in-context learning with retrieval-augmented generation (RAG) to handle tasks beyond the predefined toolset. Experiments demonstrate that an LLM using our framework can successfully perform complex tasks, from building a simple house to querying and editing existing IFC data. Our framework is released as open-source to encourage research in LLM-driven BIM design and provide a foundation for AI-assisted modeling workflows. Our code is available at https://show2instruct.github.io/mcp4ifc/.

Paper Structure

This paper contains 26 sections, 10 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. LLMs as Tool-Using Agents
  4. LLMs for BIM and 3D Design
  5. BIM Creation and Modification
  6. BIM Scene Querying
  7. LLMs for Spatial Understanding
  8. Methodology
  9. System Architecture and Workflow
  10. Toolset Implementation
  11. Scene Query and Context Tools
  12. Predefined IFC Element Creation tools
  13. Dynamic Code generation tools
  14. Experiments and Results
  15. Scene Querying
  16. ...and 11 more sections

Figures (10)

  • Figure 1: Overall system architecture of our MCP4IFC framework
  • Figure 2: Workflow of a simple wall creation request through the MCP4IFC framework.
  • Figure 3: Example IFC building elements generated through predefined tool functions. These functions also support a variety of parametric input arguments.
  • Figure 4: Duplex House
  • Figure 5: Questions and correctly given IFC-based answers by Sonnet 4.5 on the ARC Duplex using MCP4IFC.
  • ...and 5 more figures