How to use the CloudCompare agent skill: install and run it

What the skill makes the agent do

An agent skill is a folder of instructions and tools that an AI agent loads when a task matches it. This one wraps CloudCompare in a command-line interface called cli-anything-cloudcompare, so the agent can run CloudCompare operations itself.

The skill describes 41 commands across 9 groups. These cover project management, session control, point cloud operations (subsample, crop, filter, segment, normals, analyze), mesh operations, distance computation (C2C and C2M), transformations (ICP and matrix), export to LAS, LAZ, PLY, PCD, OBJ, STL and E57, and an interactive REPL.

• Create and inspect project files
• Load scans and list the clouds in a project
• Subsample, crop and filter noise from point clouds
• Compute surface normals
• Measure distances between clouds (C2C, C2M) and run ICP alignment
• Convert and export to common 3D formats

Install the skill from the ZIP or the Claude apps

Option one: download the skill ZIP from this page and unzip it into ~/.claude/skills/ for all your projects, or into a project's .claude/skills/ folder for that project only.

Option two: in the Claude apps, open Settings, go to Capabilities, then Skills, and upload the ZIP there.

The command-line tool itself is published on pip. If you want to install it directly, the skill's documented command is shown below.

Check the requirements before you start

You need Python 3.10 or newer and CloudCompare installed on the same machine. The tool wraps CloudCompare and will not work without it. On Linux, the documented route is Flatpak; on macOS and Windows, download CloudCompare from cloudcompare.org.

The skill notes it was tested with CloudCompare on Linux via Flatpak. The Linux install command from the source is shown below.

• Python 3.10 or newer
• CloudCompare installed (Flatpak on Linux, or cloudcompare.org for macOS and Windows)

Learn the global options

Global options must go before the subcommand. The two you will use most are --project, which points to your project JSON file, and --json, which makes output machine-readable so the agent can parse it.

Use the same pattern for every command: global options first, then the group and command, then its arguments.

• -p, --project TEXT: path to the project JSON file
• --json: output results as JSON (useful for agents)

Run example commands for projects and clouds

Projects are JSON files that track the clouds and meshes you load and the history of operations. Create one, add a scan, and list what the project holds. The subsample example below keeps 100000 points at random and writes the result to a LAS file.

Cloud commands take a CLOUD_INDEX, a 0-based number you can find with cloud list. Most accept --add-to-project to register the output back into the project.

• Filter noise with filter-sor (defaults use 6 neighbours and a 1.0 std ratio; both can be changed with --nb-points and --std-ratio)
• Crop to a box with cloud crop, adding --outside to keep points outside the box instead
• Compute normals with cloud normals, using --level from 1 to 10

Trigger the skill with plain requests

You do not need to type commands yourself. Once the skill is installed, describe the result you want and name the files. The agent chooses the matching commands. Be specific about file paths, the operation, and where the output should go.

Good requests state the input file, the step, and the output. For example: ask the agent to subsample a LAS scan at a 0.05 m spacing and save the result as a new LAS file, or to remove isolated noise from a scan and export it as PLY. Check the agent's commands and output paths the first time you use a new kind of request.

• Subsample /data/scan.las to 100000 random points and save it as /data/sub.las
• Remove isolated noise from the roof scan with statistical outlier removal and export to PLY
• Crop the scan to a 10 by 10 by 5 metre box and save the points outside it
• Convert /data/cloud.pcd to LAS

Safety notes: scripts and data

The skill runs scripts on your machine, since it calls the CloudCompare command-line tool. The listing says it does not use the network. Even so, read the commands the agent proposes before it runs them on important files.

Do not send confidential or sensitive scan data through an agent with network access. The skill documentation does not describe what data, if any, leaves the machine when an agent is involved, so treat that as unknown and keep sensitive scans away from it unless you have checked.

• Keep copies of original scans; output commands write new files and the session undo command does not delete output files, but a bad overwrite is still possible
• Review the agent's commands and output paths before running on important data

Tips for reliable results

Keep one project file per job so the history stays clear. Use session history to review recent operations and session save to write the project state to disk. If an operation is wrong, session undo removes the last entry from history but leaves output files in place.

Use --json when you want output the agent or your own script can parse. Set default export formats with session set-format so you do not repeat them on every command.

This guide covers the commands shown in the first part of the skill file. The skill lists more commands for ICP, distance computation, meshes and exports, and the full list is in the SKILL.md file, so check it before relying on a command not covered here.

• Use session set-format to set default cloud and mesh export formats (for example LAS or PLY for clouds, OBJ or STL for meshes)
• Use session history -n 5 to see the last five operations