What Is the Revit–Claude AI Connector?

This project is an open source bridge that lets Claude AI talk directly to Autodesk Revit.

Once installed, you can open a conversation in Claude Desktop and ask things like:

"How many walls are in this model?""List all the families used on Level 2.""Export the structural elements to CSV."

Claude reads your question, picks the right Revit tool, runs it inside your live Revit session, and answers you — all without you writing a single line of code.

---

AI Assistance in Revit

Working with Revit daily? You've probably hit that moment where a repetitive task makes you think "there has to be a better way." Here's how the automation options stack up:

📊 Dynamo — The visual programming powerhouse. Drag nodes, connect wires, see your logic flow. Perfect for BIM managers who want to build reusable workflows without deep coding knowledge. Great for geometric operations and data manipulation. The catch? Complex logic can turn into spaghetti nodes fast.

🐍 pyRevit — Python meets Revit. Write custom tools that live right in your ribbon. Speed is excellent since it runs native C# under the hood. Trade-off: requires Python programming skills.

⚡ Revit API (C#) — The fastest, most powerful option with direct access to everything Revit can do. Trade-off: steep learning curve, you often need to build your own UI, and plugins typically serve one specialized purpose.

🤖 AI Connector (e.g. Claude + MCP) — The newest player, and the one that shines where the others struggle.

• Natural language commands with the flexibility to chain tools and automate multi-step tasks.
• The real differentiator isn't just getting data out of Revit: it's what happens next. An AI connector keeps going. It can summarize findings, flag anomalies, generate a formatted report, produce a CSV, draft an email to your team, or answer follow-up questions, all in the same conversation.
• Trade-off: slower than native commands. Best for complex one-off operations, cross-discipline analysis, exploration, and anywhere the presentation of results matters as much as the results themselves.

---

The Two Moving Parts

The connector is made of exactly two pieces that talk to each other over the local network.

Claude Desktop
     │
     │  (MCP — stdio JSON-RPC)
     ▼
MCP Server  (Node.js  ·  index.js)
     │
     │  (HTTP  ·  127.0.0.1:5578)
     ▼
Revit Add-in  (C#  ·  RevitClaudeBridge.dll)
     │
     │  (Revit API)
     ▼
Your Revit Model

Piece	Language	Role
MCP Server (server/index.js)	Node.js	Speaks MCP to Claude; proxies requests to the add-in
Revit Bridge (RevitClaudeBridge.dll)	C#	Runs inside Revit; executes tools against the open model

The two pieces are deliberately separate so that the MCP Server stays simple and stateless, while all the Revit-specific logic lives in C# where the Revit API is available.

How a Request Flows

When you type a message in Claude Desktop, here is exactly what happens:

1. Claude Desktop → MCP Server — Claude decides which tool to call and sends a JSON-RPC request to the MCP Server over stdio (standard input/output of the server process).
2. MCP Server → Revit Add-in — The MCP Server forwards the tool call as an HTTP POST to 127.0.0.1:5578, the local endpoint the Revit add-in is listening on.
3. Revit Add-in → Revit API — The add-in receives the request, runs the corresponding C# logic against the live Revit model, and collects the results.
4. Revit Add-in → MCP Server — The add-in returns the results as a JSON HTTP response.
5. MCP Server → Claude Desktop — The MCP Server wraps the response in a JSON-RPC reply and sends it back to Claude over stdio.
6. Claude Desktop → You — Claude reads the results and composes a natural-language answer in the chat window.

The communication between the MCP Server and the Revit add-in is entirely local (127.0.0.1). However, your messages and the tool results are sent to Anthropic's servers by Claude Desktop in order to generate a response — the same as any normal Claude conversation.