# An Early Look at a qt-fortran GUI for ForCAD

> Source: <https://fortran-lang.discourse.group/t/an-early-look-at-a-qt-fortran-gui-for-forcad/11318#post_8>
> Published: 2026-11-18 23:32:10+00:00

**Recommendation:** Expose model-building as a scriptable API and let the agent drive it through MCP, as Astra and Fable did with Blender's Python API, so each step stays inspectable and iterable.
More in the playbook "Designing agent systems": <https://wpnews.pro/playbooks/agent-architecture>

Perfect. I have tested hooking up Astra and Fable to Blender via an MCP, so it can control it and view things. And then I just gave it photos of a chess rook and a bishop and without fail it can generate the 3D model, which in Blender is only the surface model. It creates a Python script that uses Blender’s Python API to build the mesh from the ground up, and then it can iterate on it based on my feedback (for example to cleanup some corner). I asked it to do pure quad mesh, and it did it.

I tried to learn Blender a few times and I can do some stuff and understand the basics, but I find it much easier and faster to just ask the agent to do everything for me.

So I was thinking a similar approach can work for volume meshes also, not just surface. The tool that I used for this in the past is Cubit: [https://cubit.sandia.gov/](https://cubit.sandia.gov/), and I would actually also create a script, and use Cubit in that mode — exactly as AI uses Blender now via the Python script. The script is superior since you know exactly all the steps that go into the model and can iterate on any one step, even before AI. And now with AI it becomes easy.

In your ForCAD Studio I think all that is needed is to create an API to create the model itself, which is simply ForCAD itself, the Fortran API should be good enough. Then you need to connect the GUI Studio to the agent, for example via MCP, and allow the agent to control the GUI, so that it can view the model, change how it is rendered (wireframe, surface mesh, perhaps various slices through it, as needed), etc.

Possibly controlling the GUI is not needed, perhaps just doing all the options how to render it from Fortran and then saving a png image and giving that to the AI agent is enough. I don’t know if there are pros/cons, if one is faster or more robust. If you figure it out, please let me know!

Either way, I think this will work to create any 3D hexahedral mesh for simulations, and iterate on it and make it perfect.
