Skip to content
Biminent
Revit guides

How to use AI in Revit

There are two practical ways to use AI in Revit today. You can ask a chat assistant to write scripts that you run yourself, or connect an assistant such as Claude directly to your open model through MCP, so it can read the model and, with your approval, change it.

1. Let a chat assistant write the script

Describe the task to Claude or ChatGPT and ask for a Dynamo Python node, a pyRevit script or C# for the Revit API. This works well for one-off jobs such as parameter clean-ups, exports and selections.

  • Assistants sometimes invent API methods or mix up Revit versions. Name your version and check unfamiliar calls against the API documentation.
  • The assistant cannot see your model, so describe categories, parameter names and units exactly.
  • Run new scripts on a copy of the model first.

2. Connect an assistant to Revit through MCP

MCP, the Model Context Protocol, is an open standard that lets an AI assistant call tools in other applications. For Revit, an add-in running inside Revit exposes tools such as “find elements” or “set parameter”, and an MCP server passes the assistant’s requests to it. The assistant then works on your live model instead of on your description of it.

The main options today:

  • Autodesk’s Revit Public MCP Server is a tech preview for Revit 2027, installed as an extension from your Autodesk Account. It started read-only (queries, parameter checks, reports and view snapshots), and Autodesk has since added edits that you review before they apply.
  • Open-source servers on GitHub, many built on pyRevit, are good for prototyping your own tools. Check what each one can change in the model and how it is secured before using it on project work.
  • Biminent Revit MCP, our own connector, described below.

What AI in Revit is good at today

  • Answering questions about a model: counts, missing or inconsistent parameter values, element lists.
  • Bulk parameter edits you would otherwise do in a schedule.
  • Reports and checks before issue.
  • Writing and running one-off scripts against the actual model.

It is weaker at design modelling and at anything nobody checks. Treat its changes like a junior colleague’s: preview, review, then commit, and keep a backup of the model.

What this means for Dynamo: Is Dynamo still relevant in the age of AI?