Structure Lab v0.1.0 - Minecraft 26.2 / NeoForge

Structure Lab v0.1.0 — milestone 1

Generate Minecraft NBT structures through MCP, replace them in a dedicated superflat world, and automatically inspect labeled screenshots rendered by Minecraft.

Downloads

• structurelab-26.2-0.1.0.jar — install this into a Minecraft Java 26.2 NeoForge instance.
• structurelab_mcp-0.1.0-py3-none-any.whl — the separate local MCP server and Python NBT authoring library.
• SHA256SUMS.txt — checksums for both artifacts.

Installation

1. In Modrinth, create or select a NeoForge 26.2 instance and stop it.
2. Use Content → Upload files to install the JAR, then launch Minecraft.
3. Create a world using World Type: Structure Lab.
4. Install Python 3.11+ and the MCP wheel in a virtual environment.
5. Register the Python stdio server with your MCP client, pointing --connection at <instance>/structurelab/connection.json.

Full instructions:

• Install the mod
• Install and configure MCP
• Tool reference and authoring examples

Included

• Structure Lab superflat preset, local authenticated bridge, and MCP stdio server.
• JSON operations and a Python library for NBT generation, plus file import.
• Named view-pad metadata and sidecar camera definitions.
• Automatic sequential 1080p screenshots with labels, coordinates, angles, and revision identity.
• Replacement cleanup, previous-success screenshot retention until a new success, and clean NBT/datapack export.
• Furnished example house and scripts exercising real MCP tool calls.

Compatibility and limits

Requires Minecraft Java 26.2, Java 25 for the game, NeoForge 26.2.0.77+ for 26.2, and Python 3.11+ for MCP. Tested on Windows with NeoForge 26.2.0.79 and Python 3.13. The MCP process and graphical Minecraft client run on the same machine.

This is the first milestone. One active structure is supported, with a 2,000,000-position bounding-volume cap and 4,000,000-position cleanup-box cap. Horizontal dimensions may reach 1024, but height at the default origin is limited to 380. Up to 1024 angles are accepted by the schema; five-angle batches were tested. Large-scale performance, entity templates, multi-palette structures, RCON/dedicated-server operation, ad hoc recapture, simulation validation, and automatic failed-job pruning are deferred.

The build and five Python tests passed. The live MCP integration verified generation, replacement, five 1080p captures, invalid-input handling, and export. A separate unmodded 26.2 instance loaded the exported datapack successfully. Verification details.