Skip to main content
Glama
Sign In
deenesjakoruzh
cloudcompare-mcp

cloudcompare-mcp

by yufeioptimal
OverviewSchemaSourceRelated ServersDiscussionsAdmin
Python
Local

cloudcompare-mcp

Cross-platform Model Context Protocol (MCP) server for CloudCompare — lets AI assistants (Claude, etc.) process 3D point clouds and meshes via natural language.

Features

Native tools (no CloudCompare required)

Tool

Description

read_cloud_metadata

Parse a cloud and return point count, bounding box, extent, density, RGB/intensity/normals presence

visualize_cloud

Render top / front / side views + metadata panel as a base64 PNG the model can see directly

CloudCompare tools (requires CloudCompare installation)

Tool

Description

get_cloudcompare_info

Check installation & version

load_cloud_info

Inspect file stats via CloudCompare

subsample

Reduce density — random / spatial / octree

compute_cloud_to_cloud_distances

C2C nearest-neighbour distances

compute_cloud_to_mesh_distances

C2M signed distances

icp_registration

Align two clouds with ICP

compute_normals

Estimate surface normals

filter_by_scalar_field

Threshold points by scalar value

statistical_outlier_removal

Remove noise with SOR filter

merge_clouds

Merge multiple clouds into one

convert_format

Convert between LAS/LAZ, PLY, PCD, XYZ, E57, OBJ…

run_cloudcompare_command

Escape hatch for arbitrary CLI commands

How visualize_cloud works

visualize_cloud reads the point cloud natively in Python, renders a 4-panel figure, and returns an ImageContent (base64 PNG) alongside a JSON description. The model can see the image directly — no display or CloudCompare needed.

┌─────────────────┬─────────────────┐
│   Top  (XY)     │   Front  (XZ)   │
│                 │                 │
├─────────────────┼─────────────────┤
│   Side  (YZ)    │  Metadata stats │
│                 │  (pts, bbox,    │
│                 │   density, …)   │
└─────────────────┴─────────────────┘

Color modes: height (viridis Z gradient, default) · rgb (stored RGB) · intensity (plasma).

Related MCP server: FreeCAD MCP Server

Requirements

  • Python ≥ 3.10

  • uv (recommended) or pip

  • CloudCompare ≥ 2.12 — download (only for CloudCompare tools)

Python dependencies installed automatically: numpy, matplotlib, laspy[lazrs], plyfile.

Installation

Quickstart with uvx (no install needed)

uvx cloudcompare-mcp

Install locally

pip install cloudcompare-mcp
cloudcompare-mcp

CloudCompare binary detection

The server looks for CloudCompare in this order:

  1. CLOUDCOMPARE_PATH environment variable

  2. System PATH (cloudcompare / CloudCompare)

  3. Platform default locations:

Platform

Default path

macOS

/Applications/CloudCompare.app/Contents/MacOS/CloudCompare

Windows

C:\Program Files\CloudCompare\cloudcompare.exe

Linux

/usr/bin/cloudcompare

Set CLOUDCOMPARE_PATH to override:

export CLOUDCOMPARE_PATH="/opt/custom/cloudcompare"

MCP client configuration

Claude Desktop (claude_desktop_config.json)

{
  "mcpServers": {
    "cloudcompare": {
      "command": "uvx",
      "args": ["cloudcompare-mcp"]
    }
  }
}

Claude Code (~/.claude/settings.json)

{
  "mcpServers": {
    "cloudcompare": {
      "command": "uvx",
      "args": ["cloudcompare-mcp"]
    }
  }
}

With a custom binary path:

{
  "mcpServers": {
    "cloudcompare": {
      "command": "uvx",
      "args": ["cloudcompare-mcp"],
      "env": {
        "CLOUDCOMPARE_PATH": "/path/to/cloudcompare"
      }
    }
  }
}

Usage example

Once configured in Claude Desktop or Claude Code:

"Load my scan.las file and subsample it spatially to 5 cm, then remove statistical outliers."

Claude will call the appropriate tools in sequence and report results.

Supported file formats

LAS · LAZ · PLY · PCD · XYZ · ASC · TXT · E57 · OBJ · BIN · SHP

License

MIT

Available Tools

12 tools
compute_cloud_to_cloud_distancesA

Compute nearest-neighbour distances from every point in the 'compared' cloud to the 'reference' cloud (C2C distance). Returns distance statistics and saves the labelled cloud.

Inputs: reference_path (required), compared_path (required), output_path (required), max_distance.

compute_cloud_to_mesh_distancesA

Compute signed distances from a point cloud to a reference mesh (C2M distance). Useful for comparing a scan to a CAD model.

Inputs: mesh_path (required), cloud_path (required), output_path (required).

compute_normalsA

Estimate surface normals for a point cloud using a local neighbourhood. Normals are required for many downstream operations (Poisson reconstruction, etc.).

Inputs: input_path (required), output_path (required), mode, radius, +1 more.

convert_formatA

Convert a point cloud or mesh file to a different format. CloudCompare infers the target format from the output file extension (e.g. .las, .laz, .ply, .pcd, .xyz, .asc, .e57, .obj).

Inputs: input_path (required), output_path (required).

filter_by_scalar_fieldA

Keep only points whose scalar-field value falls within [min_val, max_val]. Useful for height, intensity, or distance thresholding.

Inputs: input_path (required), output_path (required), min_val (required), max_val (required).

get_cloudcompare_infoA

Check if CloudCompare is installed and return its version and path. Call this first to confirm the tool is available.

Inputs: no named parameters.

icp_registrationB

Register a 'data' cloud onto a 'model' cloud using the Iterative Closest Point (ICP) algorithm. Returns the transformation matrix and final RMS error.

Inputs: model_path (required), data_path (required), output_path (required), overlap, +2 more.

load_cloud_infoA

Load a point cloud or mesh file and return basic statistics: number of points, bounding box, available scalar fields, etc. Supports LAS/LAZ, PLY, PCD, E57, XYZ, ASC, BIN, SHP, OBJ, and more.

Inputs: file_path (required).

merge_cloudsB

Merge two or more point cloud files into a single output file.

Inputs: input_paths (required), output_path (required).

run_cloudcompare_commandA

Run an arbitrary CloudCompare CLI command for advanced or unsupported operations. Do NOT include the executable path or -SILENT flag — they are added automatically. Example args: ["-O", "/path/cloud.las", "-AUTO_SAVE", "OFF"]

Inputs: args (required), working_directory.

statistical_outlier_removalA

Remove statistical outliers by analysing the distance distribution to k nearest neighbours. Points farther than (mean + nSigma × std) from their neighbours are removed.

Inputs: input_path (required), output_path (required), knn, n_sigma.

subsampleA

Reduce the density of a point cloud using random, spatial, or octree subsampling. Output is saved to the specified path.

Inputs: input_path (required), output_path (required), parameter (required), method.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Deploy Server
    One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

    TDQS

    A3.9/5.0

    Scored across 12 tools

    Disambiguation5/5

    Each tool targets a distinct operation (e.g., distance computation, filtering, registration, conversion), with clear differentiation even between similar tools like cloud-to-cloud and cloud-to-mesh distances.

    Naming Consistency4/5

    Most tools follow a `verb_noun` pattern with underscores, but 'icp_registration' and 'statistical_outlier_removal' deviate slightly by starting with a noun/adjective instead of a verb.

    Tool Count5/5

    With 12 tools covering loading, filtering, registration, distance analysis, and format conversion, the set is well-scoped for point cloud processing without being overwhelming or too sparse.

    Completeness5/5

    The tool set covers essential point cloud operations comprehensively, and the inclusion of an arbitrary command execution tool ('run_cloudcompare_command') fills any potential gaps, ensuring no dead ends.

    Resources

    Related MCP Connectors

    Related MCP Servers