What Happens When You Give an AI Access to Your Revit Model

Last Sunday I did something I’ve never done before. I opened Revit, opened Claude, and started asking questions about my model. Not writing C# code. Not building a filtered element collector. Just… asking.

“How many live/work units are in this building?”

And it told me. Fifteen.

“Change the name of the live/work units to loft units.”

And it did. Or at least it tried (more on that in a second).

This was possible thanks to an MCP server, and I think it’s going to change how we interact with Revit.

What Is an MCP Server?

MCP stands for Model Context Protocol. In simple terms, an MCP server sits between an AI like Claude and a piece of software like Revit. It acts as a translator, providing the AI with a set of tools to read from and write to your model.

So instead of writing code every time you want to query your model or make a change, you describe what you want in plain English. The AI figures out which tools to use and how to use them.

It’s like working with an experienced Revit user. Instead of specifying every step, you can simply describe the result you want—like saying, “rename all the plumbing fixtures”—and they handle the rest.

Why I Built My Own

There are already open-source MCP servers for Revit. I went to look at them and, honestly, I couldn’t figure out how to get started. They were confusing, and I couldn’t understand how to implement them.

So I took a different approach. I’ve been using Claude Code a lot lately, and I thought: Why not just build my own? I sat down on a Sunday morning, worked with Claude Code, and by the afternoon, I had a working MCP server connected to Revit.

That alone says something important about where we are right now. Building a bridge between an AI and Revit used to be a massive undertaking. Now it’s a weekend project.

What It Actually Looks Like

I had the Snowdon Towers model open in Revit. I launched Claude Desktop (you need the desktop version because the MCP server runs locally on your machine), and I started asking questions.

“What do you know about the current Revit model?”

Claude pulled up the project name, model size, key element types, and the fact that there were six linked models. All of that information came directly from the model through the MCP server.

“Tell me the room names on Level 2.”

It went through the server, queried the model, and returned the full list.

Then I pushed it further. I asked it to set a parameter value on those fifteen live/work units. This was a tool I had just added and hadn’t fully tested yet. And sure enough, it didn’t work perfectly. But that’s part of the process. The point is that the architecture for doing this kind of work already exists. The tools just need refinement.

The Real Insight: Your Own Toolbox

Here’s where it gets interesting. When you build an MCP server, you define a set of tools that the AI can use. My server currently has about 15 tools, like “get project information,” “create floor plan,” “get elements,” and “set parameter.”

If I asked Claude to move a wall one foot to the right, it couldn’t do it. Not because the concept is impossible, but because I haven’t given it a tool for that yet.

And that’s actually the powerful part. You get to define the toolbox.

Imagine a structural engineer building out a set of tools specific to their workflow. Tools for creating grids, placing columns, and running code checks. Or an MEP engineer with tools for analyzing systems, checking connections, and verifying sizing. Each discipline defines what matters to them, and the AI handles the rest.

You’re not coding every step of every workflow anymore. You’re building a handful of purposeful tools and then letting the AI do what it’s good at: understanding intent, choosing the right tool, and filling in the gaps.

Why This Matters

We’ve been interacting with Revit the same way for twenty years. Menus, ribbon panels, dialog boxes, and, for the more adventurous among us, C# code. MCP servers introduce something fundamentally different: a conversational interface to your model.

That doesn’t mean the Revit UI goes away. It means you have another option. One that’s particularly powerful for tasks that are tedious to do manually but hard to justify writing a full add-in for. The kind of thing where you’d normally think “I could automate this, but it would take longer to write the code than to just do it by hand.”

With an MCP server and a few well-defined tools, that calculus changes. The AI handles the logic and decision-making. You just describe what you want.

What’s Next

I’m planning to open-source this MCP server so others can try it, extend it, and build their own tools on top of it. I’ll also be sharing more about how to define custom tools for your own workflows, because that’s where the real value lies. Not in the server itself, but in the discipline-specific toolboxes that people build with it.

This feels like one of those moments where the way we work is about to shift. Not overnight, and not for everyone at once. But the pieces are falling into place. And honestly, talking to my Revit model on a Sunday morning felt like a glimpse of what’s coming.

I’ll have more to share soon. Stay tuned.