Harlan Brumm
If you’ve been watching the AI tooling space over the past year, you’ve probably noticed a lot of noise around MCP and what it means for how we work with tools like Revit. Community-built servers are popping up everywhere, and the enthusiasm is genuinely exciting. But one question keeps coming up: who do I trust?
That’s exactly the question we set out to answer with the Revit Public MCP Server.
A quick note before we dive in
This release is a Tech Preview, which means it’s available now for you to use and explore, but we’re actively developing it, gathering feedback, and iterating. It can be used in your workflow today, but you should expect things to evolve. We’d love to hear what’s working and what isn’t.
What Is MCP, Anyway?
MCP stands for Model Context Protocol, an open standard that lets AI assistants connect to external tools and data sources in a structured way. Think of it as a USB standard for AI: instead of every AI tool having to build its own custom integration with every application, MCP gives AI agents a common language to plug in and communicate. For Revit, that means an AI assistant like Claude can ask your model direct questions and get answers back, without you having to copy-paste data, export spreadsheets, or describe your model manually. The AI connects to Revit through the MCP server, and Revit responds.
What Is It, and Why Does It Matter?
The Revit Public MCP Server is Autodesk’s official, supported MCP server for Revit 2027. It gives AI assistants like Claude a direct and structured way to read and understand your Revit models. Think of it as a trusted bridge between your BIM data and the AI tools, you’re already starting to use day-to-day.
What are Autodesk MCP Servers? Autodesk MCP Servers are trusted tools made for agent-driven AI workflows in Design and Make. Built to Autodesk’s realiability standards, they act as safe bridges, letting AI assistants connect with Autodesk software, data, and tasks in a secure and dependable way.
Getting Started
The Revit Public MCP Server ships as a separate addon for Revit 2027, and you’ll need Revit 2027 installed to use it. You can download it from accounts.autodesk.com under your Revit 2027 entitlements.
Once installed, Claude Desktop or Cursor is configured for you automatically! But for those technical folks, the configuration looks like this:
{
“mcpServers”: {
“revit”: {
“command”: “C:\\Program Files\\Autodesk\\Revit 2027 MCP Server Technical Preview\\RevitMCPServer.exe”
}
}
}
The server communicates over stdio, which means it works with any LLM client that supports the MCP specification. Claude Desktop or Cursor is a great starting point, but the same configuration pattern applies to other MCP-compatible clients and tools.
What Can It Do Today?
The current release focuses on read-only access to your Revit model. That’s intentional. Before we open additional capabilities, we want to make sure the foundation is solid: queries are fast, results are accurate, and the experience is one you can trust.
Here’s exactly what’s in the box. Seven tools, each doing one thing well:
Get Running Revit Instances
The required first step in any session. This tool detects which Revit instances are currently running on your machine and returns the process ID and open document name for each. Other tools take that process ID as its input, but the AI knows which model it’s talking to, even if you have multiple projects open at once.
Query Model
Query Model is the workhorse. This tool searches your model using multi-criteria filtering: category, family name, element name, level, bounding box, and parameter values with support for equals, contains, starts with, ends with, greater than, and less than comparisons. You can scope the search to the current view, a specific view, or the entire model. Results come back in two parts: a list of element IDs you can pass directly to other tools, and an analysis summary showing element counts broken down by category, level, and more. That analysis section is especially useful for narrowing down follow-up queries without having to pull full element data first.
Get Element Data
Once you have element IDs from a query, this tool retrieves the details. You control what comes back: basic info (name, category, family, type, level), element class, bounding box coordinates, key parameters, or the full set of all instance and type parameters. It’s how your AI assistant goes from “I found 47 structural columns” to telling you what each one is made of, where it sits, and what its parameters say.
Select Elements
Bridges AI analysis and your active Revit session. Pass it a list of element IDs and those elements become selected in Revit immediately, ready for you to inspect, tag, schedule, or modify. A direct handoff from AI output back to your hands.
Zoom to Elements
Takes things one step further than selecting. Rather than just selecting elements by ID, this tool zooms in and focuses the current view directly on them. When the AI has found something specific, this is how it brings you right to it on screen.
Open View
Activates a specific view in Revit by its ID. Useful when the AI needs to navigate you to the right floor plan, section, or 3D view as part of a larger workflow.
Export View
Exports views to image files (PNG, JPG, BMP, or TIFF), separate PDF files, or schedule data as CSV, with three export modes: a specific list of views by ID, the current active view, or the visible region as it appears on screen. Output goes to a path you specify, or defaults to your Documents folder. This is what enables AI-assisted documentation, rendering pipelines, and reporting — the AI can drive exports as one step in a larger task.
Together, these seven tools give your AI assistant a more complete picture of your model and a set of safe, read-only actions to work with, without ever touching your data in a destructive way.
The Strategy: A Base Layer You Can Build On
Here’s where it gets interesting for the AEC community. We see the Revit Public MCP Server as a foundation, not a ceiling.
Because it’s an official, tested server, it becomes a reliable base that you, your firm, or third-party developers can extend. You can run it alongside other Revit MCP servers, including community-built tools, firm-specific automation, and specialty plugins, and let each layer do what it does best. You’re not locked in, and developers don’t have to reinvent the wheel just to get reliable read access to a Revit model.
This is about choice and flexibility on the desktop. Your AI stack should work the way your practice works: mix-and-match, composable, and firmly in your control.
The Same Engine That Powers Autodesk Assistant
Around the same time as this Tech Preview, Autodesk is also launching the Autodesk Assistant Tech Preview, a new AI assistant built directly into Autodesk products. If you haven’t seen it yet, the Autodesk Assistant is designed to help you get answers, navigate tools, and work through tasks without leaving your application. It’s early days for the Assistant too, but the direction is clear: AI that understands your design context, not just generic prompts.
What the MCP server and the Assistant have in common is the underlying Revit model intelligence. The same technology that lets the Assistant understand your Revit context is what we’re exposing through the MCP server for use in your own AI desktop workflows.
The practical difference right now is that the Autodesk Assistant operates in a more controlled environment and can do more within that context out of the box. The MCP server gives you something different: the ability to bring that same intelligence into whichever AI client you choose, on your terms. They’re complementary.
What’s Next
We’re actively working on expanding these capabilities. Future tools, giving AI assistants the ability to create, modify, and manage Revit elements, are part of where we’re headed. We plan to release those through a dedicated write server, keeping read and write concerns cleanly separated so you can control what your AI assistant is allowed to do in your model.
The goal is straightforward: a trusted, official source of Revit MCP tools that users can operate for their workflows and that developers can build on and extend without starting from scratch every time.
We’ll have more to share as things mature. In the meantime, grab Revit 2027, our MCP server addon, set it up for your environment, and start asking your models better questions.
