The MCP Server for Revit - Python: An Alternative to the Autodesk MCP Server Tech Preview
pyRevitLabs doesn't ship an official "pyRevit MCP." What people usually mean is the community project MCP Server for Revit – Python. It's a pyRevit extension that connects Revit to Claude or any other MCP client. It uses pyRevit's Routes module as a bridge between Revit and LLMs, and it's meant as a simple template you can build on to give an AI access to your Revit models.
How it works: main.py is the MCP server that talks to Claude. When a tool is called, it turns the call into an HTTP request and forwards it to Revit. The pyRevit Routes piece is a small REST API that runs inside Revit, executes Revit API code, and returns JSON.
Tools currently working:
- Status and model info: checking that the connection is live, pulling general information about the model, and listing levels with their elevations
- Views: exporting a view as an image, listing all exportable views, and reading info about the active view and the elements visible in it
- Families: placing a family instance at a given location, and listing family types and categories
- Visualization: color-coding elements by parameter value, clearing those overrides, and listing a category's parameters with sample values
- Code execution: running IronPython directly inside Revit. Through this, the AI can do almost anything the API allows.
- Documents: finding installed Revit versions, launching Revit, opening, closing and saving documents, and syncing with central
Still pending: getting the current selection, creating line- and surface-based elements such as walls and floors, deleting and modifying elements, and tagging walls.
Safety features: You pair it inside Revit through MCP > Bridge > Connect, and any request that changes the model or runs Python shows you its payload in Revit before it goes ahead. The bridge only accepts local connections, so remote or LAN access is rejected.
Status: The project is being prepared for Autodesk Marketplace review but isn't certified yet. The authors describe it as more of a demonstration than a finished product. There's also a larger fork supporting Revit 2024 through 2027 with 48 tools that cover design, editing, analysis, clash detection, MEP, interop and documentation.
Comparing the Autodesk MCP Server Tech Preview to the MCP Server for Revit - Python
These two servers take opposite approaches.
- Autodesk's is official and locked down by design.
- The community Python server is more open, can edit the model, and lets you add your own tools.
Comparison
Autodesk Tech Preview | MCP Server for Revit – Python | |
Who makes it | Autodesk (official) | Community, open source (MIT) |
Revit versions | 2027.2 only | Whatever pyRevit supports; a fork covers 2024–2027 |
Depends on | Nothing extra; installs from your Autodesk account | pyRevit with the Routes server enabled, plus Python/uv |
Searching the model | Strong: multi-criteria filtering with parameter comparisons and summaries | Basic: model info, levels, elements in the current view, family lists |
Element data | Detailed and configurable | Category parameters with sample values |
Selecting and zooming in live Revit | Yes | Selection is still pending |
Exporting | Images, PDF, schedules to CSV | View images only |
Editing the model | Only in the Write and Experimental configurations | Places families, applies color overrides, and runs arbitrary IronPython |
Document control | Not in the launch toolset | Launches Revit, opens, saves and closes files, syncs with central |
Custom tools | Meant to sit alongside other servers, not to be extended | Built to be forked: two files plus two registration lines per tool |
Safety model | Separate read and write servers | Pairing in Revit, a payload review before edits or code runs, and localhost-only access |
Which to use
- For questions, audits and QC ("how many fire-rated doors on Level 3 have no mark?"), use Autodesk's server. Its query tool is far more capable, it doesn't need pyRevit, and it's the supported option.
- For editing, document handling or firm-specific tools, the Python server is more flexible. The tradeoff is that much of that power comes from
execute_revit_code, which means the AI writes and runs code against your model. Keep the review prompt on and test on a copy first. - You can run both. Autodesk's launch post says the official server is meant to run alongside community and firm-built MCP servers. A common setup is Autodesk's server for reading and the Python server, or your own fork of it, for the writing it doesn't cover yet.