Private Paper dev server on a deep superflat world, tuned for safe creative build tests.
Detailed Project Profile / Minecraft Builder
Active local prototype
Minecraft Java
Mineflayer
Qwen assisted
A local bot learning to build real places in Minecraft.
Manna Minecraft Builder is a private local Minecraft Java project where Grayson can ask Manna to build something and a Mineflayer bot turns a compact plan into validated Minecraft commands.
The goal is not random structure spam. The long arc is a builder that can generate coherent castles, villages, roads, dungeons, adventure hooks, and eventually whole lore-aware settlements that feel like places worth exploring.
What Exists Now
The prototype is already past a toy command demo
The project is still private and local, but the working path is real: Java server, bot, planning boundary, deterministic execution, live testing, and visual critique.
Mineflayer bot with chat commands, preview, confirm, cancel, progress, settings, and undo.
Validated command generation with block-name checks, fill splitting, and runtime budgets.
Reference-world flythroughs and live visual review become generator-quality rules.
Current Capabilities
What Grayson can test in the world
- Gatehouse, walls, towers, paths, banners, courtyard props, dungeon threshold
- Better gables, armory and smithing identity, tower floors, portcullis detail
- Still needs slimmer tower silhouettes and stronger composition variance
- Road, plaza, well, smithy, guard post, farmhouse, tavern, farms, market props
- Rich variant adds wall/gate relationships, awnings, lanterns, gardens, tree clusters
- Next quality unlock is collision/layout reservations before scaling larger
- Spawn staging, path, ruined gate landmark, tower, underground challenge, reward room
- Built from lessons in adventure maps, route gates, reveal moments, and reward staging
- Future routes should connect placed landmarks to quest text and world context
Design Lessons Learned
The generator is being taught composition, not just block placement
The strongest recent shift was architectural: stop overfixing one castle or village stamp and build reusable build-generation logic.
Witchcraft and Wizardry, Stonehill Castle, Old Fallen Castle, Across The Time II, and personal survival builds became learning material for route, facade, ruin, and settlement grammar.
StudiedGrayson's survival kingdom photos point toward mountain-valley kingdoms, walls, roads, bridges, fields, dramatic roofs, warm lighting, turquoise fantasy accents, and terrain integration.
CapturedRoads should connect, facades need identity, landmarks need reveal moments, rewards need staging, and live feedback should become reusable generator grammar first.
Operating rulePrompts like house, watchtower, castle, village, dungeon, road, bridge, and mountain site need routing and composition layers, not one repeated village pattern.
Next turnGrand Vision
What the project is actually aiming for
Supported prompt shapes
Manna should understand the difference between "simple house on a mountain", "one big watchtower with a cone roof", "village", "castle", and "dungeon" before scaling into larger colony generation.
Lore-aware settlements
Grayson should eventually be able to ask for a colony or kingdom and get a coherent place with roads, buildings, dungeons, rewards, palettes, adventure hooks, and sensible world context.
Boundaries
What the project should not claim yet
This is a private local creative/dev system, not a survival-safe public-server bot.
AI can choose compact specs, but deterministic validated code executes the build.
The goal is authored generation grammar, not blindly pasting giant copied structures.
Different prompts need different composition plans, placement rules, and module families.