This workshop establishes a new paradigm for AI-assisted architectural design, moving beyond static image generation into fully adaptive, parametric system design. The process is built around the Model Context Protocol, enabling Large Language Models like ChatGPT Astra and Claude Fable to operate architectural software directly.
The workflow progresses through a definitive 4-stage hybrid methodology: Extraction, User Checkpoint, Modeling, and Verification. Crucially, your approach will prioritize "mapping before prompting", translating briefs into explicit rule sets and structured JSON files before the agent touches the model. You will act as the architectural editor, managing context engineering and token budgets, while the AI acts as your draftsman, rigorously applying rules to generate native geometry and BIM data.
Scope
Throughout the workshop, students will focus on integrating AI agents into established architectural pipelines, using a mid-rise office block on a constrained plot as a testable typology.
You will:
- Install, configure, and triage Rhino MCP and Revit MCP for end-to-end operational use.
- Drive Rhino to automate massing generation, layer discipline, and group definitions.
- Drive Revit to query data, audit parameters, create native families, and automate sheet exports.
- Drive Grasshopper to build your first fully parametric system operated entirely by LLMs.
- Benchmark frontier models (Astra vs. Fable) based on accuracy, instruction adherence, recovery, and token economics.
Program
- Setup & The State of the Race: Install Rhino/Revit MCP, configure servers, and explore frontier models, agentic vs. generative AI, and MCP as the AEC connective layer.
- Methodology, Tokens & Cache: Map the parametric approach, build the project context JSON, and manage token budgets and caching strategies.
- Modeling with Rhino MCP: Drive massing generation, enforce layer discipline, and execute modeling with LLMs through guided case studies.
- Modeling with Revit MCP & Benchmarking: Audit live models, extract data, model parametric families, and automate project documentation. Compare ChatGPT Astra against Claude Fable.
- Modeling with Grasshopper: Transition from static LLM modeling to fully parametric system design, parameter definition, and Python scripting via LLMs.
- Conclusions & Security: Address permissions, backups, and prompt-injection risks on live models, and discuss the future of the architectural profession.
.jpg&w=3840&q=75)
