You can use the Revit MCP to use LLMs to control Revit in three ways with increasing capabilities:
- Autodesk Assistant — use the Autodesk Assistant chat window built into the Revit interface.
- Revit MCP — use an LLM tool (Claude, ChatGPT (CODEX), or others) that is connected to the Revit MCP and can use the features available there.
- The Revit MCP is currently available in three versions:
- MCP Server Read-Tools Technical Preview
- MCP Server Write-Tools Technical Preview
- MCP Server Experimental-Tools Technical Preview
- Download and install the Experimental-Tools version for the most complete set of tools. Installing the Revit MCP Server
- Revit MCP + Dynamo (Python) — use an LLM tool and allow it to generate Python code that uses the Revit API to access capabilities beyond the features available in the Revit MCP.
- The current implementation uses Dynamo to run this Python code and modify the Revit project model.
How is the Workflow Different?
As you move down the list to more capable tools, additional software must be installed, and the workflow for making changes involves more steps:
- Autodesk Assistant — enter prompts directly into the chat window in Revit.
- Revit MCP — enter prompts into the LLM tool interface, and it communicates with the Revit MCP to make changes.
- Revit MCP + Dynamo (Python) — enter prompts into the LLM tool interface, and when it cannot execute the prompt directly, it offers to write a Python script that can be run in Dynamo to make the changes in Revit.
Which Method Should I Use?
The AI prompts shared on this site indicate which options are needed for running each prompt.
Prompt | Autodesk
Assistant | Revit MCP | Revit MCP + Dynamo |
Set the Vertical Mullion Interior Type property to None for wall type Curtain Wall: Storefront. | 🟢 | 🔵 | |
Create a new curtain wall type with every other curtain panel set a Solid type to create a checkerboard pattern | 🟣 |
- Green Dot = can be run in Autodesk Assistant
- Blue Dot = can be run in an LLM using Revit MCP
- Purple Dot = can be run in an LLM using Revit MCP and will require a Python script to be generated