Setting up SketchUp MCP means getting two pieces talking: a Ruby extension inside SketchUp, and an MCP server your AI client can reach. The open-source route is free and takes about twenty minutes if nothing goes wrong. This page covers it, and the route where the app does all of it for you.
Before the steps, the shape of it — this is what makes the failures readable later.
A Ruby extension loads inside SketchUp and opens a local socket. A separate Python process is the MCP server: your AI client launches it, and it forwards typed tool calls to that socket. So there are three things that can be wrong — the extension is not loaded, the server is not running, or the client cannot start the server. Almost every setup problem is one of those three.
The community servers are distributed as Python packages run through uvx. On macOS:
brew install uv
On Windows, use the installer from the uv project. Confirm it is on your PATH:
uvx --version
If that command is not found, stop here and fix it. A large share of later failures are really this one surfacing in a confusing way — see spawn uvx ENOENT.
Download the project's .rbz file. In SketchUp go to Window → Extension Manager, click Install Extension, and select the .rbz. Restart SketchUp.
Some projects also expose a settings panel under Plugins where Ruby evaluation has to be enabled before the extension will accept commands. If the extension installed but nothing responds, check there first.
Add the server to your MCP client's configuration file. For a client using the standard format:
{
"mcpServers": {
"sketchup": {
"command": "uvx",
"args": ["sketchup-mcp"]
}
}
}
The file must be valid JSON. A trailing comma or a smart quote pasted from a web page will make the client fail silently with no server listed — this is the single most common setup mistake.
Fully quit and reopen your AI client; most do not re-read the config file while running. The SketchUp tools should now appear in the client's tool list. Open a model and ask for something small — "how many groups are in this model?" — before trying anything that creates geometry.
The extension and the server talk over a local port. Which one depends on the project: 9876 and 8080 are both in use across the community implementations. If something else on your machine already holds that port, the extension will load and the server will start, and nothing will work. See the port section.
ZirenAI does the whole of Route 1 for you, per application. It detects which modelling software is installed, and sets each one up on its own — the extension, the runtime it needs, and the wiring between them. There is no package runner to install, no JSON to edit and no AI client to register, because the client is the app.
If you are comfortable with a terminal and already pay for an AI client, the open-source route adds nothing to your bill and does the core job. It is a real option and we would not pretend otherwise — note the already pay, though: the server is free, the modelling is not.
The app route is worth paying for if twenty minutes of PATH debugging is not how you want to spend an evening, if you would rather not manage a second AI subscription, or if you want the connection checked for you every time instead of discovering mid-task that the extension quietly stopped loading after a SketchUp update.
Every failure we have seen falls into one of about eight buckets, each with a recognisable error string. They are collected with fixes on one page.
Troubleshooting →