Guide · Agent skill

How to use the Blender Scene CLI agent skill (cli-anything-blender)

Updated 2026-10-29 · written with AI from the project's own documentation and checked against our catalogue data

This guide explains the cli-anything-blender agent skill, which lets an AI agent edit 3D scenes in Blender through a command-line tool instead of the graphical interface. It is aimed at artists, designers and developers who have not used agent skills before. By the end you will know what the skill needs, how to install it, how to phrase requests so the agent triggers it, and what safety points to check first.

Blender Scene CLIHKUDS/CLI-Anything · Apache-2.0 Download the skill

Quick start

1Confirm Python and Blender versions. Check that Python is 3.10 or newer and Blender is 4.2 or newer on your system.
2Install the package. Run pip install cli-anything-blender in your terminal.
3Install the skill folder. Unzip the downloaded skill into ~/.claude/skills/ or a project's .claude/skills/, or upload the ZIP under Settings → Capabilities → Skills in the Claude apps.
4Test the CLI. Run cli-anything-blender --help and create a test scene file with scene new using an absolute path.
5Ask for a scene in plain language. Describe the objects, materials, lights or camera you want and name Blender in your request.
6Capture a preview. Ask for a quick preview so the agent produces hero.png and workbench.png and reports their paths.
7Review and undo if needed. Open the images, and ask the agent to undo or show session history if a change is wrong.

What the skill makes the agent do

The skill teaches an agent to work with Blender, a free 3D modelling and rendering program, through a stateful command-line interface. The agent can create a scene project file, add 3D primitive objects, assign materials, add cameras and lights, set animation keyframes, and render images or previews.

Every change is tracked, so the agent can undo and redo operations (up to 50 levels, according to the source). Commands can also return JSON, which lets the agent read results back and check its own work.

Install the skill

Download the skill ZIP from this page and unzip it into one of these locations: ~/.claude/skills/ for your personal skills, or a project's .claude/skills/ folder if you want the skill limited to one project. Alternatively, in the Claude apps, go to Settings → Capabilities → Skills and upload the ZIP.

The skill package itself is installed with pip. The source lists this command for the CLI tool that the skill calls:

Python 3.10 or newer is required.
Blender 4.2 or newer must be installed on your system; the skill cannot render without it.
The skill's package is named cli-anything-blender on pip.

pip install cli-anything-blender

Check the tool works before asking the agent

Before you rely on the skill, confirm the command runs. The source shows a help command and a command that creates a new scene project file.

Run the help command to confirm the tool is on your path.
Create a test project file in a folder you do not mind changing.
Use absolute paths for all file operations, as the source requires for agents.

cli-anything-blender --help
cli-anything-blender scene new -o scene.blend-cli.json

How to trigger the skill with natural requests

You do not need to type commands yourself. Describe the scene in plain language and name Blender so the agent knows which skill to use. Ask for visual checks when you want to see results, since the source recommends preview renders over judging scenes from JSON alone.

Good requests name the objects, the material or light, and whether you want a preview. Vague requests such as 'make something nice' give the agent little to work with.

'Using the Blender CLI skill, create a new scene called desk.blend-cli.json with a plane and a cube.'
'Add a metallic material with roughness 0.3 to the cube, then capture a quick preview.'
'Add a camera and a sun light, set the frame range to 1 to 120, and list the keyframes on the cube.'
'Undo the last change and show me the session history.'

Main command groups

The CLI is organised into groups. The agent picks the right one for each step. You can ask for any of these in a request.

scene: new, open, save, info, profiles, json
object: add, remove, duplicate, transform, set, list, get
material: create, assign, set, list, get
modifier, camera, light and animation (keyframe, frame-range, fps and related commands)
render: settings, info, presets, execute, script
preview: recipes, capture, latest and live sessions
session: status, undo, redo, history

Preview renders and visual checks

For visual verification, the source describes a preview capture. For the recipe quick, the typical artifacts are hero.png (from Eevee preview rendering) and workbench.png (from Blender Workbench rendering). The agent reads these image files; the JSON only references their paths.

Live preview sessions can also be started, which save a session.json head, immutable bundle folders and a trajectory.json history of commands and previews.

Ask for a preview capture after major changes.
Ask the agent to report the artifact file paths so you can open the images yourself.
The cli-hub previews commands listed in the source can inspect, convert to HTML, watch or open existing previews.

cli-anything-blender --json --project scene.blend-cli.json preview capture --recipe quick

Safety notes before you run it

Rendering runs Blender-generated scripts on your machine, and the source says it may overwrite project files. Do not point the skill at sensitive or production scene files. Work on copies, and keep backups of anything you care about.

The skill runs scripts but does not use the network, according to the listed requirements. Review the commands the agent proposes, especially any that save or render over existing files.

Use copies of your .blend work and a dedicated project folder.
Always pass absolute paths for project and output files.
Check return codes and stderr when a command fails, as the source's agent rules advise.

What people use it for

Prototyping a simple product or set layout by adding primitives, materials and a camera step by step.
Setting up basic lighting and a camera, then checking the look with preview renders.
Adding keyframes and setting the frame range and FPS for a short animation test.
Building a scripted pipeline where an agent renders previews and reads JSON results for automated visual checks.

Frequently asked questions

Is the cli-anything-blender skill free?

The source lists the skill under the Apache-2.0 licence and describes Blender as a free program. Check the licence file in the repository for the full terms before you use it in a commercial project.

What do I need to use the Blender CLI agent skill?

You need Python 3.10 or newer, Blender 4.2 or newer installed on your system, and the pip package cli-anything-blender. The skill cannot render without Blender.

Where do I put an agent skill so Claude can use it?

Unzip the skill into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder. In the Claude apps you can also upload the ZIP under Settings → Capabilities → Skills.

Does the skill use the internet?

The listed requirements say it does not use the network, but it does run scripts. Review what the agent runs, and avoid running it on sensitive scene files.

Why does the agent need JSON output?

JSON output lets the agent read results such as scene info and file paths in a structured form. The source says every command can return machine-readable JSON with the --json flag.

What if a command fails or the agent undoes the wrong change?

Non-zero return codes signal errors, and the agent should read stderr for the message. Use the session commands undo, redo and history to step back through changes.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

This skill suits artists, designers and developers who want an AI agent to build and check Blender scenes step by step, especially when they can use preview renders to verify results. Skip it if you do not have Blender 4.2 or newer, or if you need to work on sensitive or production scene files, since rendering runs generated scripts on your machine and may overwrite project files.

Download the skill