When a chat assistant answers a question from an open Revit model, or changes a parameter in it, there is often an MCP server doing the connecting. Model Context Protocol is a standard way for an AI client, such as Claude or another chat app, to call tools that a separate program exposes. A Revit MCP server is a Revit add-in that exposes tools like reading element parameters, selecting by category, setting values or creating sheets. The AI client you already use can then work on the model you have open.
Search for one and you'll get a mix of paid products, open-source projects and Autodesk previews, several of them with nearly the same name. Here is who offers what, and what actually differs once you try to put one to work.
Revit MCP server commercial vendors
LUBIMKA sells RevitMCP through the Autodesk App Store as a subscription product. Its help page lists support for Revit 2024 through 2027, and the App Store listing is the clearest way to confirm you have the right publisher.
Nonica offers its own product at revitmcp.com. It's a separate company with a separate product, despite the similar name. That name collision causes real confusion. “Revit MCP” is a description, so a tutorial, forum thread or setup command that mentions it could mean either of these products, an open-source project or Autodesk's own work. Check the publisher before you copy an install step from a guide.
Autodesk is now part of this picture too. Alongside the Revit 2027.3 and 2027.4 releases, it published public Revit MCP add-ons as a Technical Preview, with Read, Write and Experimental extensions, plus DynamoMCP, which in Dynamo 4.2.1 with Revit 2027.3 can build and run Dynamo graphs. As a preview it comes with Autodesk's preview terms, which differ from a paid product with a vendor standing behind it. Separately, Autodesk's built-in Revit Assistant in Revit 2027 and later handles help, model queries and model changes inside Revit, with no outside AI client required.
There are also open-source Revit MCP projects, generally free to use and maintained by their contributors. Our Revit Model Context Protocol guide explains how these connections work.
What actually differs between them
The biggest practical difference is who keeps the connector working. Revit updates change the APIs that add-ins depend on. With a paid product, keeping up is the vendor's job. With an Autodesk preview, Autodesk sets the pace on its own terms. With open source, it's the maintainers, or your own team if you need a fix sooner.
Version support follows from that. LUBIMKA's help page lists 2024–2027, Autodesk's preview arrived with the 2027 point releases, and open-source projects cover whichever releases their maintainers have built and tested. Check the listed versions against the releases your active projects actually use, especially if long-running jobs are still on Revit 2023.
You also have to supply the rest of the setup. An MCP server needs an AI client and an account with an AI provider, and whether it works depends on the whole combination: the Revit build, the connector release, the client, the connection method and the account settings. A product that “supports MCP” hasn't necessarily been tested with your client. When you get a working setup, write down the exact combination so the next workstation can match it.
Then there's whether the tools fit the task. Take a request like “update a door schedule.” It could mean changing the schedule's fields and filters, editing parameter values on the doors themselves, or changing how the schedule is formatted. Those are three different operations, and a connector may expose one of them, generate code for another and leave the third to you. Tool counts are a poor substitute for checking this. A small toolset that covers your actual work is more useful than a long catalog missing the one operation you need.
Where your model data goes
A Revit MCP plugin runs on your machine, which leads many people to assume the whole workflow stays local. It usually doesn't. Whatever the tools return, whether parameter values, element lists or file contents, goes to the AI client and from there to the AI provider, sometimes through a vendor relay. A local plugin doesn't make the workflow offline.
The product and client documentation should say what is sent and which account controls apply. Confirm those details with whoever handles deployment for your firm. Knowing the path also helps when something breaks, because you can tell whether the problem sits in the Revit plugin, the connection or the assistant.
A connector or a complete workspace
A connector suits teams that want to choose their own AI client and run the surrounding setup themselves. Some teams would rather not assemble those pieces, especially when the work involves more than the model: a spec PDF, a spreadsheet, a marked-up screenshot from a review.
ArchiLabs Desktop packages all of that into one Windows application. It edits native models in Revit 2023 through 2027 and works with AutoCAD content, and you attach PDFs, spreadsheets, screenshots and screen recordings to the same conversation as the model, so you don't have to assemble a separate AI client and connector yourself. It starts at $19 a month plus usage credits.
Understand Revit Model Context Protocol, Autodesk's Read and Write servers, Claude and OpenAI connection options, and where ArchiLabs Desktop fits.