The instructions for adding it to Claude are based on the old version of Claude Desktop. They moved that functionality to Customize. Shaw is showing this also in their messages yesterday. I saw it uses port 10500 by default. I tried adding the json myself:
You don’t have Extensions under a “Desktop App” subheading? I am using the latest version of Claude Desktop on Win/Mac if you see different to my screenshot please show me.
The file downloaded is “connector.mcpb” but connectors are managed under Connectors. You should download extension.mcpb
Also, the installation tries to connect to router/osx-x64/rhino-mcp-router which is incorrect on an Intel Mac. So I tried changing it to port 10500 as shown in my previous message, but it says Server disconnected.
It’s okay if it doesn’t yet support Rhino3D on Intel Macs. I know it requires having a Mac to test. But the instructions should tell the user and the installer should stop and warn the user.
This is true only for ones in the Anthropic store, which this one is not currently.
As I said above you’ll need to bypass the router which is not built for Intel macs. Open Rhino, Run MCPStart and press enter for 10500. Then paste that URL into your AI Agent and ask it to add this as an MCP Server. That will probably work.
I understand that the router is not yet available for Intel Macs. You said to try port 10500 directly. You are saying that this extension DOES work on Intel Macs?
The router is an intermediary between Rhino and the AI Agent, its goal is to make things simpler and more extensible, but it is not strictly necessary. You could (probably) connect directly between Rhino and the AI Agent without the router.
I would love to be able to automate boring stuff, taking a 3d model and placing on layouts and adding dimensions and stuff is pretty time consuming in rhino.
We have 3 pipelines that can be explored.
Static make2d, which I dislike because it makes stuff way harder to edit afterwards.
Dynamic make2d, that is, section tools, that is making stuff a bit easier but I dislike that we don’t have ways of separating the files between 3d and 2d so the file gets poluted in case you want to use it as a block or with worksessions.
And viewports, that is my prefered way to not generate extra stuff inside the files but sometimes wonky stuff happens, like layours not preserving their show/hide status between sessions or some render aberrations when using differente view styles.
So being able to ask the model to create the layouts for the object in question would be very nice.
I usually have my layers separated by material, by construction step. But I work with wildly different geometries everytime. So being able to ask "Hey, take construction step 1 and create 3 pages of layouts with standar projection (or with this, this and this projection) one with metalic only, one with plastic parts only and the third one with them assembled, and add name tags and general dimensions.
its the kind of stuff that is not hard for a human to do, but in the end it costs some time that could be used in other more valuable tasks.
Last night I almost finished a new command I’m developing with Claude connected to Visual Studio Code. So far, everything seems to be working correctly: the output file is an .rhp, and I’m currently optimizing it because I want the commands to have a floating options window, similar to the new Patch command in Rhino WIP.
The crazy part is that, to my surprise, it actually works!
I’m now creating the icons for the commands, and I also asked Claude to generate a dedicated toolbar to organize them.
Unfortunately, my Claude subscription only allows a certain amount of work before I have to wait for the session limits to reset, so I have to take breaks between development sessions. In the meantime, I’m continuing to refine and improve the commands.
The most absurd part of all this? I know absolutely nothing about programming!
I basically just explain what I want the command to do, how it should behave, and which options it should have… and Claude writes all the code for me. It honestly feels like having a personal programmer working at my side.
@Olav1 If you ask Claude to handle the installation, it will do it for you. At some point, it will tell you that it needs to install something called uv/uvx. Just give it permission to proceed with all the installation steps, and in the end Claude will install the Rhino MCP plugin for you.
You need to install Visual Studio Code and then install the Claude AI desktop app. After that, Claude will guide you through the rest of the setup.
Copy and paste both links, then ask Claude to carefully read the procedures in each one before proceeding.
These are two different tools, both valid but with different purposes:
rhinomcp (McNeel)
lamcp (gramaziokohler)
Made by
McNeel (creators of Rhino)
University research lab
Focus
Rhino 3D in general
Grasshopper-specific
What it does
Creates/modifies 3D objects in Rhino
Reads/modifies Grasshopper components and sliders in real time
Official
Yes
No
rhinomcp is better suited if you want Claude to create geometry in Rhino (e.g. “create a sphere”, “move this object”).
lamcp is better suited if you want to interact with Grasshopper definitions (e.g. “change the value of this slider”, “add a component”).
Has anyone connected Claude to MCP in WiP? I have wasted 3 hours trying to do so, letting Claude to review the process, altering json and some other s***. Every time Claude was trying to establish connection it was opening Rhino 8.