Last week, a longtime 3D printing journalist did something that would have sounded like science fiction two years ago: he opened a CAD program he had never used, typed a sentence into ChatGPT, and watched a bookshelf design itself on his screen. Then a five-blade propeller. No CAD training, no tutorials, no menus. Just a conversation.
The experiment, published by Fabbaloo's Kerry Stevenson on July 28, used a free, open-source combination that anyone can set up today: FreeCAD plus a small plugin that speaks something called MCP, the Model Context Protocol. Here is what actually happened, what it means for makers and engineering buyers, and where the humans still earn their keep.
What just happened with FreeCAD and ChatGPT?
MCP is a standardized way for AI systems to control external software. Think of it as a common language: the tool publishes a list of commands it understands, and the AI calls them. A community-built MCP plugin for FreeCAD (the actively maintained one is by GitHub user neka-nat) installs in minutes. Copy a folder into FreeCAD's Mod directory, restart, flip on the new MCP workbench, and connect it to ChatGPT's local tool runner. From that point, the AI can create documents, add and edit objects, run code inside FreeCAD, and even kick off FEM analysis.
Stevenson's tests escalated quickly. A simple cube worked on the first try. Then he asked for a three-shelf bookshelf, one meter wide and 30 cm deep with 30 cm tall shelves. ChatGPT built it almost instantly, assumed a sensible 18 mm board thickness, and even added a thin back panel he never asked for, because it knew what a bookshelf should look like.
Can I just describe a part to AI and have it 3D printed? I don't know CAD at all.
Yes, for simple parts you can now describe what you want in plain English and have an AI generate a real, editable CAD model, but for functional parts you still want an experienced human to check dimensions, tolerances, and printability before anything hits a build plate. The Fabbaloo experiment proves the first half of that sentence, and one small detail proves the second half: the propeller test. Stevenson asked for five blades, a 10 cm hub, and five 5 mm bolt holes. The propeller looked right on screen, but the bolt holes were silently missing. The AI fixed them when asked, and it fixed them well, but nobody would have caught the problem without looking closely at the model.
That is the state of AI CAD in one story. Impressive speed, real geometry, and quiet omissions that only show up when someone who knows what a bolt circle is inspects the file. A missing hole in a desk ornament is a shrug. A missing hole in a bracket that mounts to your machine is a returned shipment.
I have a broken part and no CAD file. Can someone recreate it for me?
Yes. This is exactly what 3D scanning and reverse engineering are for: we scan your physical part, rebuild it as a clean CAD model, correct any wear or damage, and print a new one, no original drawings required. AI text prompts work when you can describe a part from scratch. But most real-world jobs start with an object, not a description: a discontinued bracket, a cracked housing, a vintage car trim piece nobody makes anymore. Describing that in words to a chatbot loses everything that matters, like the exact hole spacing and that weird draft angle the original mold left behind. A scanner captures it all in one pass. If that is your situation, our 3D scanning and reverse engineering service handles the whole chain from physical object to fresh part in hand.
Where this is heading, and where the humans stay
Stevenson's conclusion matches what we see in the shop: novices can now get good results on simple parts, and future models will keep climbing the complexity ladder. That is great news for everyone. More people designing parts means more parts that need printing, in real materials, with real quality control.
Because here is the part the demo does not show: a CAD model is the start of the job, not the end. Material selection, print orientation, tolerances after cooling, post-processing, and shipping are where a part succeeds or fails. Whether your model came from twenty years of SolidWorks experience or from a chat window last night, we can take it from file to finished part through our Build and Ship service. Upload the file, tell us what the part needs to survive, and we handle the rest here in Rockville, MD.
So go ahead and try the FreeCAD MCP setup this weekend. It is free, it is genuinely fun, and when your AI-designed part is ready to become a physical object, or when it turns out the AI forgot your bolt holes, email us at info@dcadditivepros.com.
Frequently asked questions
Can I use ChatGPT to make a 3D model for printing?
Yes. With the free FreeCAD MCP plugin, ChatGPT can create and edit real CAD models from plain-English prompts. It works well for simple parts like boxes, shelves, and basic shapes, though complex or precision features still need human review.
Is AI-generated CAD accurate enough for functional parts?
Sometimes, but verify everything. In published tests the AI produced correct overall geometry yet silently omitted specified features like bolt holes. For load-bearing or precision-fit parts, have an experienced designer or print shop check the model first.
I need a replacement for a discontinued part. Can someone 3D print one?
Yes. A shop with 3D scanning can capture your existing part, rebuild the CAD model, and print a replacement, even with no original drawings. This works for brackets, housings, knobs, trim pieces, and most other discontinued components.
Do I need to buy expensive CAD software to get a custom part made?
No. FreeCAD is completely free, AI assistants can now help you model in it, and a print service can take your file (or your physical object) from there. You can go from idea to shipped part without buying any software.