I’ve been building a SketchUp extension for cabinetmaking and furniture work for my own needs, and before I sink more evenings into it, I’d like to check what other people would need the mos so it’s helpfull to most.
The idea: generate cabinets and drawers directly in SketchUp with real hardware (Blum, Hettich…), correctly placed, and automatically output the dimensioned 2D shop drawing with the drilling positions plus the list of part numbers to order. Basically go from model to shop-ready without re-checking everything by hand.
Here’s where it’s at — a few cabinets with generated drawers:
You have a great idea. I’m actually working on a closet maker for my own use as well.
Are you building it with AI? I use a hybrid workflow with ChatGPT 5.6 and Opus 5. I’ve found that ChatGPT 5.6 is especially good at modeling smaller details like hinges and handles.
One thing that has worked well for me is modeling one part at a time. It gives the AI a much clearer focus and usually leads to better results. For references, IKEA has been very helpful since they provide detailed dimensions for many of their products.
Thanks ! Yes, i am using claude AI atm. I’m not generating anything through IA appart from panels though, since i want real and accurate models added i have to do them piece by piece. Yes for good result i found it’s better to give him the full plan as mush as possible and then go into details afterwards so he connects all the dots on his own and make good choices.
Claude (through Claude Code) isn’t bad at modeling, either. I asked it to write an extension that launches a local server it can connect to—allowing it to run whatever Ruby code it likes within SketchUp—and it manages to both analyze my geometry and recreate it.
Here is a preview of what it draw when I ask him to draw test cases for a new cavity detection feature :
I mostly use it to improve my drawing tools, but essentially, there’s nothing stopping it from drawing cabinet boxes.
The only issue is that the “thinking” time is quite long, and token consumption can be significant.
Claude is definitely great. I use it when I want a safer, more predictable coding experience. GPT-5.6 is more likely to change things outside the requested scope, but for micro-modeling, such as hinges, door handles, and other fine details, GPT-5.6 (Ultra) performs exceptionally well.
I see, not bad ! The plan would be to not have to use the AI for the plugin itself, more realiable and less energy focused indeed, but that’s a good idea I wonder how i can implement him do to the job for me regarding all those real life components with realiable measurements i’ll be adding though.
Yeah, i have looked it up allready but thanks for your input. My goal is to get close to this level of customisation, with proper hardware positionning and plans. It’s gonna be hard to get to that point, but you guys would be interested by something like that ? Would it save you a lot of time daily ? I know it would in my own working process.
It always save time to automate this kind of task. The hardest part is finding the right balance between a good level of customization—including offbeat scenarios—and a rigorous structure.
In the current DEV version of OpenCutList I try to add 2 tools that permits to add connectors and fittings with their machinings. However, even though it works quite well for common use cases, the system easily fails when faced with atypical cabinet configurations.
Question on this - How are you having it model? Are you having it generate the elements with Codex using a Ruby Script? Or is there some kind of MCP connection for GPT that I’m not aware of?
If it can help someone, here is the “Bridge” extension I used to connect Claude Code to my active SketchUp. It launches a local server and Claude can run any script on SketchUp by it self. Maybe more complete than a MCP serveur as it exposes all what ruby can do on current open model .
@boris.beaulant Wow, that’s great thanks for sharing ! I don’t know how it’s gonna help me yet but it probably will. My plugin so far is designed to create the cabinets, panels, doors on itself, and use a library i’ve made to import accessories according to real life rules i’ve implemented beforehand. Dunno if i can speed up the process of adding real life accessories with the help of IA yet but i’ll definitely try. I’ll probably share a few videos soon once i’m happy with the development.
Maybe I gave Claude a bit too much credit. Here’s an example. I extracted some 3D models from the IKEA Planner and converted them into Ruby code. This is what ChatGPT 5.6 Ultra produced compared to Opus 5 (highest). The difference is huge.
I’m not sure what the legal position is on this, but I’ve seen people on X discussing and experimenting with extracting 3D models from online planners. It’s an interesting trend, though it’s probably worth checking the licensing and terms of use before experimenting with it.
Speaking of MCP servers, I was able to replicate the entire IKEA PAX Planner UI and UX. Now all I need to do is bring in the models, and I’ll have my own PAX designer inside SketchUp.
How accurate and reliable is you ikea planner so far in terms of measurments ? I might follow up with a post of my own work soon when i’m more advanced on the project .
SketchUp is still needed because that is where the architecture and the entire project come together. The cabinet interface only makes one part of the workflow easier. The cabinets still need to be placed inside the full model so the project can be presented to the client, rendered, or viewed through a 360-degree walkthrough.
For MCP, I used a Chrome MCP server with a browser extension and computer-use connection to extract the interface, models, and other parts of the planner.
Currently, I’m working on a feature for Kitchen Crafter that can build an entire kitchen automatically. You would simply highlight a line, and the plugin would generate a recommended kitchen layout. From there, you could replace, edit, or move the cabinets around, saving a lot of time.
The measurements are working well so far, but that’s not the hard part. The real challenge is bringing all the models in after converting them to Ruby code. I’m using Luna Max for this project, and it can run all day while consuming very little of my usage quota.
I wonder how i can implement him do to the job for me regarding all those real life components with realiable measurements i’ll be adding though.
.
