airirang/airirang-builder

AIrirang Builder is a tool that turns a single chat instruction into a complete Minecraft structure. Instead of asking an AI agent to place blocks one by one, you describe what you want and the tool produces a datapack that builds the whole thing in seconds inside the game. The README is explicit that this is not an official Minecraft product and is not approved by or associated with Mojang or Microsoft. The core idea is a pipeline that runs end to end in one step. The tool reads a 3D mesh file in .obj format, converts it into Minecraft voxels, compresses those voxels into Minecraft /fill commands using a method called greedy meshing, and then wraps the result into a Java Edition datapack you can drop into a world. The flagship tool is called quick-build, and it does the full chain in a single call. Once the datapack is installed, you run /reload followed by /function airirang:<id> in Minecraft and the structure appears. The project connects to AI clients through an open standard called MCP, short for Model Context Protocol. You register the tool with Claude Desktop by adding a short snippet to a config file, and after a restart the client sees seven new tools: list-presets, voxelize-preset, voxelize-custom, generate-mcfunction, execute-build, quick-build, and list-builds. A standalone command-line interface is also available for users who do not want to go through an AI client. The package bundles five free medieval models from Quaternius released under CC0, including a house, an inn, a mill, and a sawmill. You can also pass your own .obj file. Requirements are Node 20 or newer and Minecraft Java Edition between version 1.20.5 and 1.21.x. Settings let you change the namespace, pick a block override map (for example using spruce planks for wood), set a pitch correction for the mesh, and choose whether to fill the interior. The README compares the approach to a competing project that uses a bot to place each block individually, claiming faster builds, deterministic accuracy, and far fewer AI tokens per build. Compression benchmarks cited include around 95 percent reduction on a realistic building and 80 percent on a more complex lattice. The project is Apache 2.0 licensed and bundles no Mojang assets.