pyldraw3

A modern Python package for creating and manipulating LDraw format files - the standard for CAD applications that create LEGO models. It is a drop-in replacement for the unmaintained pyldraw library.

🖥️ pyldraw3-tui

Prefer an interactive interface? pyldraw3-tui is a companion terminal UI — built on Textual and pyldraw3 — for browsing the LDraw parts catalog and inspecting .ldr/.mpd model files without writing code: look up part codes, explore categories and minifig sections, inspect part metadata with colour swatches and sub-part trees, and open models to view pieces, statistics, and bill-of-materials data (read-only, it never modifies your files).

Features

• 🧱 Complete LDraw Support: Full compatibility with the LDraw standard format
• 🐍 Pythonic API: Import LEGO parts directly as Python modules
• 📦 Dynamic Library Generation: Automatically generate Python modules from LDraw libraries
• 🧭 Building Instructions: Typed STEP/ROTSTEP, LPub3D semantics, validation, manifests, and MPD/LDR snapshots
• 🔎 Reliable Inspection: Tolerant loading, structured diagnostics, exact geometry, provenance, and one-pass analysis
• 🔗 LDCad/Studio Connectivity: Typed metadata coverage, strict contacts, stable IDs, and occurrence multigraphs
• 📜 Comprehensive Guide: Jump into the quick start below, or read the published documentation

Table of Contents

• Features
• Quick Start
  • Installation
  • Setup
  • Examples
  • Basic Usage
  • Reading and Writing Model Files
  • Reliable Loading and Inspection
  • Building Instructions
  • Part Geometry Queries
  • IDE Autocompletion and Type Checking
• Requirements
• Agent Skill
• Configuration
• CLI Reference
• Development
  • Setup Development Environment
  • Development Commands
  • Documentation Site
• Architecture
  • Core Components
  • Key Classes
• Contributing
• License
• Trademarks
• Credits

Quick Start

Installation

uv add pyldraw3

Setup

Activate your virtual environment and set up the LDraw library - this will download the LDraw library and create the parts classes:

source .venv/bin/activate
ldraw download --yes
ldraw generate --yes

By default ldraw download fetches the complete LDraw release (~80 MB, everything LDraw publishes). To pin a specific dated release instead - useful for reproducible builds or a smaller download - pass --version, e.g. ldraw download --version 2018-02 --yes. Each downloaded release is cached separately, and ldraw generate builds ldraw.library.* from whichever release is currently configured (see Configuration).

Examples

Check the examples/ directory for sample scripts demonstrating various features:

# Run an example
python examples/figures.py > my_model.ldr

Each example writes LDraw text to stdout; see the Examples documentation for a description of what every script demonstrates.

Basic Usage

This package allows users to create LDraw scene descriptions using Pieces which are Parts that have a specific position and orientation. Piece.to_ldraw() and Group.to_ldraw() produce LDraw file content; str(piece) and str(group) delegate to those serializers:

from ldraw.library.colours import Light_Grey
from ldraw.library.parts.bricks import Brick1X2WithClassicSpaceLogoPattern
from ldraw.pieces import Group, Piece
from ldraw.geometry import Vector, Identity

# Create a simple model
model = Group()
Piece(
    Light_Grey,
    Vector(-10, -32, -90),
    Identity(),
    Brick1X2WithClassicSpaceLogoPattern,
    model,
)

with open("my_model.ldr", "w") as ldr_file:
    print(model, file=ldr_file)

ldraw.library.* is generated by ldraw generate and gives you every part as an importable, autocompletable Python name (as used above). If you'd rather look a part up by its catalog description or LDraw code at runtime - for example when the part name isn't known until your program runs - load the parts catalog directly instead:

from pathlib import Path

from ldraw.config import Config
from ldraw.parts import Parts

config = Config.load()
parts = Parts(Path(config.ldraw_library_path) / "ldraw" / "parts.lst")
cowboy_hat = parts.get_entry_by_description("Minifig Hat Cowboy").code  # -> "3629"
head = parts.get_entry_by_description("Minifig Head with Solid Stud").code  # -> "3626a"
brick1x1 = parts.get_entry_by_description("Brick  1 x  1").code  # -> "3005"

Both cowboy_hat and Brick1X2WithClassicSpaceLogoPattern are just LDraw part code strings, so either style can be passed as the part argument to Piece.

For new code, Piece.place offers a keyword-first constructor with sensible defaults (main colour, origin position, identity rotation):

from ldraw import Piece

piece = Piece.place("3005", colour=4)  # red 1x1 brick at the origin

The core construction types are all importable from the top-level package: Model, Piece, Group, Person, Colour, Vector, Matrix, Identity, Parts, and the read/validate/BOM helpers. Deep imports (from ldraw.pieces import Piece) keep working.

Building Models Programmatically

Model bridges piece construction and file I/O — build, query, and save:

from ldraw import Group, Model, Person, Piece, Vector

model = Model.from_pieces(
    [Piece.place("3001", colour=4)],
    name="scene.ldr",
    description="A scene",
    author="you",
)

# Figures flow in through their group
group = Group()
figure = Person(position=Vector(0, -48, 0), group=group)
figure.head(colour=14)
figure.torso(colour=4)
model.add_group(group)

# Submodels: registers the section and returns the referencing piece
wheels = Model.from_pieces([Piece.place("3005", colour=0)], name="wheels.ldr")
model.add_submodel(wheels, position=Vector(0, -24, 0))

model.find_pieces(colour=4)  # query by part and/or colour
list(model.iter_pieces())  # leaf pieces, submodels expanded
model.bill_of_materials()  # counted (part, colour) rows
model.save("scene.ldr")

Reading and Writing Model Files

read_model parses whole .ldr and .mpd files - including MPD 0 FILE / 0 NOFILE sections - into a Model you can inspect, modify, and save:

from ldraw import read_model

model = read_model("my_model.ldr")
print(model.description, model.author)

for piece in model.pieces:
    print(piece.part, piece.position)

# MPD documents: the first 0 FILE section is the root model,
# later sections are submodels resolvable from their type-1 references.
for ref in model.pieces:
    if (submodel := model.submodel_for(ref)) is not None:
        print(f"{ref.reference} -> {len(submodel.pieces)} pieces")

model.save("my_model_out.ldr")

Every parsed object (Piece, Line, Triangle, Quadrilateral, OptionalLine, Comment, MetaCommand) has a to_ldraw() method, so parsed content round-trips back to LDraw text - subfile reference casing included, byte for byte. Parse errors report the file and 1-based line number. ldraw validate exposes the same checks on the command line.

Building steps are first-class: model.add_step() appends a 0 STEP marker and model.steps returns the pieces grouped step by step. Header lines are managed through model.set_header(description=..., name=..., author=..., ldraw_org=..., license=...).

Reliable Loading and Inspection

For user-facing tools, load_model() reads, parses, and validates once. A malformed line becomes a stable, structured diagnostic while valid content around it remains available in a partial model:

from ldraw import DiagnosticCode, load_model, prepare_catalog

parts = prepare_catalog().parts
result = load_model("model.mpd", parts=parts)
for diagnostic in result.diagnostics:
    print(diagnostic.code, diagnostic.section, diagnostic.line_number)
    if diagnostic.code is DiagnosticCode.MODEL_UNKNOWN_PART:
        print(f"  unknown part: {diagnostic.offending_value}")

if result.model is not None:
    analysis = result.analyze(parts)
    print(analysis.summary.occurrence_count, analysis.bom)

Use prepare_catalog() for one cancellable setup operation, public PartsCatalog.search() for consistent multi-field search, and parts.inspect_part(code) / inspect_model(model, parts) for metadata, exact geometry, root-to-leaf occurrence provenance, and explicit failures. First-run clients can call discover_libraries(), inspect_library(), and plan_download() before changing anything. Preview rendering remains optional: render_capabilities() reports whether LDView or LeoCAD is present. The same report-oriented workflows are available from the CLI:

ldraw parts geometry 3001 --format json
ldraw inspect model.mpd --format json -o inspection.json
ldraw render model.mpd --view front --view top --backend auto

Inspection retains valid occurrences from partially malformed input and includes structured diagnostics in JSON. It exits non-zero when those diagnostics contain errors; geometry warnings remain reportable successes.

Building Instructions

Version 1.4 adds a semantic instruction layer without changing raw Model.objects, Comment, MetaCommand, or the legacy Model.steps view. Each reachable MPD model keeps its own section-local step sequence:

from ldraw import InstructionBuilder, Model, Piece, RotationMode, Vector

submodel = Model.from_pieces(
    [Piece.place("3001", colour=16)],
    name="module.ldr",
)
InstructionBuilder(submodel).step()

model = Model(name="main.ldr")
builder = InstructionBuilder(model)
model.add_submodel(submodel, colour=4, position=Vector(20, 0, 0))
builder.note("Source page 17")
builder.rotation_step(0, 90, 0, mode=RotationMode.ADDITIVE)

document = model.instruction_document()
for section in document.sections:
    for step in section.steps:
        print(section.name, step.number, len(step.added_occurrences()))

model.save("instructions.mpd")

The parser recognizes standard STEP, MLCad ROTSTEP, the supported structural/camera subset of modern and legacy LPub3D commands, and namespaced !PYLDRAW notes, highlights, and 3D arrows. Unsupported LPub3D directives remain attached to their original raw objects and round-trip losslessly.

The instruction CLI inspects and validates structure, exports deterministic JSON, and creates cumulative MPD plus flattened LDR snapshots. It intentionally does not render PDF, image, HTML, or page layout:

ldraw instructions inspect instructions.mpd --parts
ldraw instructions validate instructions.mpd --strict --max-parts 25
ldraw instructions export instructions.mpd -o instructions.json
ldraw instructions snapshots instructions.mpd --out snapshots

Part Geometry Queries

Parts can answer placement questions directly from the library's part files, resolving subfile references recursively:

from ldraw import Parts

parts = Parts.get("~/ldraw/parts.lst")

box = parts.bounding_box("3001")  # axis-aligned, in LDU
print(box.min, box.max, box.size)  # origin sits on the stud plane; +Y is down

print(parts.stud_positions("3001"))  # centres of the 8 top studs

for stud in parts.studs("3062b"):  # every stud primitive, tubes included
    print(stud.name, stud.description, stud.position, stud.is_top_stud)

for connection in parts.connections("32000"):
    print(connection.kind, connection.role, connection.position, connection.axis)

parts.geometry(code) expands drawable points exactly and returns bounds, studs, typed physical connections, and completeness diagnostics together. Connection features cover studs/receptacles, bars/clips, pins/holes, Technic axles/axle holes, hinge fingers, and tyre/rim seats. Profiles retain diameter, axial span, cross-section, friction/sliding freedom, hinge detents, source, confidence, and occupancy rather than reducing every connection to a point. bounding_box() remains as the compact compatibility helper. Stud queries expand stud group primitives down to individual stud* references; is_top_stud distinguishes upward connectors from underside tubes. Use ldraw parts geometry CODE [--format table|json] for the same query without writing Python.

Primitive-based inference is deliberately conservative. Projects that have LDCad shadow metadata can layer it on without making it a required dependency:

parts.add_connection_shadow("~/ldcad/shadow")  # directory, .zip, or .csl
parts.add_studio_metadata("studio-connectivity.json")

report = parts.connection_metadata("32000")
print(report.coverage, report.source_count, report.diagnostics)

SNAP_CYL, SNAP_CLP, SNAP_FGR, SNAP_GEN, SNAP_INCL, and SNAP_CLEAR are supported; direct feature/clear commands embedded in custom part files are consumed as inline metadata too. Parts.set_connection_overrides() provides the final, authoritative layer for custom parts. Adding or clearing either source invalidates derived geometry automatically.

Connection precedence is heuristics, primitives/official shortcuts, inline LDCad, registered shadow sources, registered Studio sources, then explicit overrides. Replacement is by stable feature ID; different IDs remain separate even when colocated. Coverage is complete for clean authoritative metadata (including an explicit clear-to-empty result), partial for recovered or primitive/heuristic-only evidence, and none only when no metadata or connector primitive exists.

Official complete wheel/tyre shortcuts provide evidence-backed pairing via parts.tyre_rim_compatibility, compatible_tyres(rim_code), and compatible_rims(tyre_code). Connections inherited by an already assembled wheel-and-tyre shortcut are marked occupied.

inspect_model(model, parts) transforms connection frames into world space. Use inspection.connection_contacts() to find interfaces already mated and inspection.snap_candidates(moving, fixed=...) to obtain ranked rigid placements. Round interfaces preserve free roll; Technic axle profiles enforce their quarter-turn cross alignment. Snap deltas are always proper rotations, so a part placed with a mirroring matrix is never reflected. inspection.connection_graphs() returns confirmed and optimistic immutable multigraphs over zero-based occurrence indices. Strict stud contacts require opposing receptacle evidence plus an oriented entry-face and penetration check; overlapping surfaces alone are not connections.

IDE Autocompletion and Type Checking

The package ships a py.typed marker, so the hand-written API is typed for mypy/pyright out of the box. For the generated ldraw.library.* modules, run:

ldraw stubs

from your project root. This writes an ldraw-stubs/ PEP 561 stub package that Pylance/pyright discover automatically (for mypy, ensure the project root is on mypy_path). Regenerate the stubs after switching library versions with ldraw download/ldraw generate, and add ldraw-stubs/ to your .gitignore. Use --out PATH to write the stubs somewhere else. Stub discovery from a project root is standard PEP 561 behavior but can vary by tool version - the stubs mirror the generated modules exactly, so pointing your checker's stub path at them always works.

Requirements

• Python 3.12+

Agent Skill

The lego-model-builder skill turns natural-language model descriptions into reusable pyldraw3 programs, verified LDraw files, preview images, and parts lists.

Download lego-model-builder.zip

See the beginner-friendly installation and usage guide for ChatGPT and Claude desktop setup, example prompts, expected files, and troubleshooting. The guide also includes a command-line installation option for technical users.

Configuration

ldraw download and ldraw generate write their settings to a YAML config file in an OS-appropriate config directory (via platformdirs). Run ldraw config to see the current values:

$ ldraw config
generated_path: /Users/you/Library/Application Support/pyldraw3/generated
ldraw_library_path: /Users/you/Library/Caches/pyldraw3/2018-02

• ldraw_library_path - the downloaded LDraw release currently in use (switch releases by re-running ldraw download --version ...)
• generated_path - where ldraw generate writes the ldraw.library.* package that you import

CLI Reference

The published docs include a standalone CLI reference.

usage: ldraw [-h] command ...

Manage LDraw libraries and create, inspect, validate, and render LDraw models.

positional arguments:
  command
    download  Download and unpack an LDraw parts library release.
    generate  Generate the ldraw.library modules from the downloaded library.
    parts     Query the parts catalog.
    validate  Validate an LDraw file (.ldr, .mpd, or .dat).
    bom       Print a bill of materials for an LDraw model file.
    inspect   Inspect exact world bounds, provenance, contacts, and gaps.
    render    Render a deterministic set of standard preview views.
    stubs     Write a type-stub package for ldraw.library into your project.
    instructions
              Inspect, validate, and export renderer-neutral instructions.
    config    Print the current configuration.
    version   Print the installed pyldraw3 version.

options:
  -h, --help  show this help message and exit

• ldraw download [--version VERSION] [--yes] - download and unpack an LDraw release (default version: complete)
• ldraw generate [--yes] [--force] - (re)generate ldraw.library.* from the currently configured release; --force regenerates even if already up to date
• ldraw parts search TERM [--limit N] - search the catalog by description or code substring (exit code 1 when nothing matches)
• ldraw parts info CODE - show a part's description, category, file path, and the generated-library import to use
• ldraw parts geometry CODE [--format table|json] - report exact expanded bounds, studs/connectors, and completeness diagnostics
• ldraw validate FILE [--strict] - lint a file: malformed lines, unknown parts and colour codes are errors; suspect matrices, legacy dithered colours, and unknown meta-commands are warnings (--strict makes warnings fail; exit code 1 on errors)
• ldraw bom FILE [--format table|csv|json] [-o OUT] - print a bill of materials counted by part and colour, submodels expanded
• ldraw inspect FILE [--format table|json] [--gap-threshold LDU] [--chronological] [--ldcad-shadow PATH] [--studio-metadata PATH] [-o OUT] - report exact occurrence geometry, provenance, typed contacts/graphs, structured diagnostics, and nearest AABB gaps
• ldraw render FILE [--view front|isometric|top] [--backend auto|ldview|leocad] [--output-dir DIR] [--overwrite] - render a safely staged named view set through the optional preview backends
• ldraw stubs [--out PATH] - write an ldraw-stubs/ PEP 561 stub package for IDE autocompletion
• ldraw instructions inspect|validate|export|snapshots ... - inspect semantic steps, validate instruction structure, export JSON, or write cumulative MPD/LDR snapshot bundles (requires a configured parts catalog)
• ldraw config - print the current configuration as YAML
• ldraw version - print the installed pyldraw3 version

Run ldraw <command> --help for a command's full option list.

Development

This project uses uv for dependency management and packaging.

Setup Development Environment

# Clone the repository
git clone https://github.com/hbmartin/pyldraw3.git
cd pyldraw3

# Install dependencies
uv sync

# Activate virtual environment
source .venv/bin/activate

# Download and set up LDraw library (fetches the latest complete release;
# pass --version 2018-02 to pin a dated release for reproducible builds)
uv run ldraw download --yes
uv run ldraw generate --yes

Development Commands

# Run tests
uv run pytest                 # All tests
uv run pytest --cov=ldraw     # With coverage (97% regression gate)
uv run pytest --integration   # Integration tests only

# Code formatting and linting
uv run ruff format .         # Format code
uv run ruff check            # Lint code
uv run ruff check --fix      # Fix linting issues

# Build package
uv build

Documentation Site

The documentation site is built with Zensical from the Markdown sources in docs/. The API reference is expanded from ldraw.__all__ at build time and rendered from type annotations and docstrings via mkdocstrings-python.

uv sync --group docs
uv run zensical serve
uv run zensical build --clean --strict

GitHub Pages must be configured to publish from GitHub Actions. If Pages is still configured to publish from a branch, GitHub will keep serving that branch instead of the Zensical workflow artifact.

Architecture

Core Components

• CLI Interface (ldraw/cli.py): Command-line interface with catalog, validation, inspection, rendering, BOM, instruction, generation, and configuration commands
• Instruction Semantics (ldraw/instructions.py): Sectioned STEP/ROTSTEP and LPub3D interpretation, authoring, inventory, and validation
• Instruction Artifacts (ldraw/instruction_artifacts.py): Deterministic JSON manifests and cumulative MPD/LDR snapshots
• Dynamic Library Generation (ldraw/generation/): Converts LDraw libraries to Python modules (with .pyi stubs)
• Import System (ldraw/imports.py): Custom meta path hook for dynamic imports

Key Classes

• Model (ldraw/model.py) - Reads and writes whole .ldr/.mpd model files
• InstructionDocument / InstructionBuilder - Read and author renderer-neutral building instructions
• Parts - Manages parts catalog and loading
• Piece - Represents individual LEGO pieces in models
• Person (ldraw/figure.py) - High-level minifigure construction
• Geometry classes - Matrix operations and 3D mathematics

Contributing

Contributions are welcome! See CONTRIBUTING.md for the fork/branch/PR workflow.

License

This project is licensed under the GNU General Public License v3.0 or later - see the license (COPYING) file for details.

pyldraw, a Python package for creating LDraw format files.
Copyright (C) 2008 David Boddie <david@boddie.org.uk>
Some parts Copyright (C) 2021 Matthieu Berthomé <matthieu@mmea.fr>
Some parts Copyright (C) 2025 Harold Martin <harold.martin@gmail.com>

Trademarks

LDraw is a trademark of the Estate of James Jessiman. LEGO is a registered trademark of the LEGO Group.

Credits

• Original Author: David Boddie
• Previous Maintainer: Matthieu Berthomé
• Current Maintainer: Harold Martin

This repository was extracted from the original Mercurial repository and modernized for current Python practices.