Construct skill
An agent describes a thing in words, and Construct checks it and builds a hologram, a 3D model with parts, a print and a kit.
- 121 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- Scripts run locally, write scene and export files and open a browser for review stills.
- The `cli.py author` command sends the description to the chosen model with your key.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/constructDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Scripts run locally, write scene and export files and open a browser for review stills.
- The `cli.py author` command sends the description to the chosen model with your key.
This is third-party code. Review the repository files before installing.
What it does
Construct is a toolkit for AI agents that turns a description of a thing into a checked scene: named parts at real sizes, joints, motion and a mark for how each fact is known (measured, inferred or guessed). The agent researches and writes the description, and Construct checks it and builds an interactive web page in lines of light in one house style, Measured Light. The same scene yields a 3D model with working joints for Blender and game engines, FreeCAD and STEP files, a print, a model kit with a booklet, and a film. The `construct_hologram/package.py` script builds the skill for Claude Code and Hermes, and `construct_hologram/server.py` runs an MCP server. Python and a browser are the only dependencies.
Who it is for. Makers, 3D artists and game developers who ask an agent to model a real thing and want to see what is measured and what is guessed.
Good fit when
- A famous building or car needs a 3D model with named parts and sizes.
- A broken part is drawn from its dimensions and photos for printing or CAD.
- A game project needs a prop with working joints for Godot, Unity or Unreal.
- A museum shows a reconstruction where visitors can tell evidence from design.
- A NetLogo run is turned into a visual scene for a lesson.
Not a fit when
- The model must serve as an exact survey for construction: the README calls the result a planning model, not a survey.
- The machine has no Python 3.10 or Chromium-family browser for review stills.
- You expect a result in minutes: masterpiece mode runs in rounds with an independent critic and can take hours.
Example request
Build a construct of my electric guitar from these photos in fast mode and show which sizes are measured and which are guessed.Limitations
Python 3.10 or newer is required, a Chromium-family browser for review stills, `ffmpeg` for video and Blender 4.2 or newer for the add-on. Quality depends on the model: for named subjects in masterpiece mode the README advises an Opus or Sol class model, and small models report their limits. The unbounded mode runs for hours, so the skill never picks it on its own. The README calls print kits a first version not yet built by a person. One NetLogo example is distributed under CC BY-NC-SA 3.0.
How to disable. Remove the skill from the agent skills folder or end the session started with `--plugin-dir`, and delete the `construct` MCP entry if added. Delete the repository clone and generated scenes, frames and exports separately.
FAQ
How do I connect it to Claude Code?
Build the skill with `python3 construct_hologram/package.py --out dist`. Then run `claude --plugin-dir dist/claude-code/construct` or copy `dist/claude-code/construct/skills/construct` into `.claude/skills/`.
Is there an MCP server?
Yes. Add a `construct` entry to the client config with the command `python3` and the path to `construct_hologram/server.py`.
How do modes differ?
`fast` makes one pass with self-review in about ten minutes. In `masterpiece` an independent critic model judges rounds, and it runs longer. The skill never picks unbounded modes by itself.
How does Construct show confidence in the data?
Each part is marked: size and shape are known separately, and guesses are shaded by how firmly they are held. Key `P` switches to that view.
Are API keys needed?
Construct itself holds no keys or network services. A key matters only if you run `cli.py author` yourself against an OpenAI-compatible endpoint.
Discussion
No questions or reviews yet.
Related
SkillsHolo Card Studio builds a collectible holographic card from four image layers, a Blender scene and a Three.js web viewer.
Blender Agent Studio builds models and scenes in Blender from a description: a .blend file, Python source and checked renders.
Blender MCP
MCP for Blender
An MCP server and Blender addon: the agent creates and edits objects, materials and scenes, pulls assets and runs Python in Blender
One line naming a machine, and Claude Code builds an interactive physics-based 3D model of it in a single HTML file
A loop skill: the agent generates a target screenshot, builds toward it, and uses a critic subagent to close the gap until they match
Skills, a CLI and an MCP server that let Claude Code or Codex make video from a brief: editing, graphics, voiceover, subtitles
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Construct (this card) | no | 121 | ||
| Blender Skills | no | 2,675 | drives Blender through the BlenderMCP add-on; Construct builds and checks the scene itself and gives Blender an export | |
| FreeCAD MCP | no | not required | 3,310 | an MCP for working inside FreeCAD; Construct yields FreeCAD and STEP as one output of a checked scene |
| OpenSCAD MCP | no | not required | 170 | renders and measures OpenSCAD parts; Construct marks the provenance of every dimension |
README in short
The README gives a who-are-you table by role, examples with animated previews, a description of fast and masterpiece modes, requirements and a per-host install table.