Heroquarium ArtGen MCP Server
Enables Claude to generate pixel art procedurally and normalize external images into a canonical palette and tile size.
README
Heroquarium ArtGen
Pixel-art asset pipeline for the game Heroquarium: procedural generation
(procgen), external-art normalization (pixelize), and an MCP server that
exposes both to Claude. See ARTGEN_BLUEPRINT.md for the full design.
Canon: generate at TILE=32, export downscaled to 16 for the current
renderer. All output snaps to the 53-colour palette in palette.json (single
source of truth). RGBA, binary alpha, light from top-left.
Layout
palette.json canonical palette (53 colours) — do not diverge
artgen/
palette.py load, redmean nearest-match, quantize, ramps
canvas.py pixel-map DSL, value noise, dither, shadow, outline, mirror
pixelize.py normalize any image/atlas -> palette + tile size
procgen/ procedural generators (M2+)
mcp/ MCP server (M5)
tools/ original game sources, kept for reference
tests/ palette + canvas acceptance tests
references/ style refs + Recraft prompts (for nngen / M7-M8)
out/ generated art (gitignored)
Setup
pip install -e '.[dev]' # runtime + pytest
pytest -q # 14 tests
pixelize
# single sprite -> 16x16 in palette
python -m artgen.pixelize in.png -o out/tree.png
# atlas: 4x4 grid, or 32px source tiles -> separate files
python -m artgen.pixelize sheet.png --grid 4 4 -o out/
python -m artgen.pixelize sheet.png --src-tile 32 -o out/
--method box|lanczos for "fat" (AI/painted) art, nearest for already-pixel art.
procgen
python -m artgen.procgen terrain --tile 32 --seed 42 --variety 3 -o out/terrain/
python -m artgen.procgen nature --tile 32 --seed 42 -o out/nature/
python -m artgen.procgen autotile --from grass --to water -o out/edges/
python -m artgen.procgen building --size L --roof red --walls stone -o out/
python -m artgen.procgen village -o out/village/
python -m artgen.procgen all -o out/
Buildings: parametric houses (size S/M/L x roof slate/red/thatch x walls stone/
timber/frame) plus a prop kit — fence, gate, well, barrel, crate, signpost,
lantern, market stall, woodpile, haystack. village writes the whole catalog +
a gallery sheet.
Terrain biomes: grass (+flowers), forest_floor, dirt, sand, stone, snow, swamp,
plus animated water frames — all tileable (seamless when repeated). Nature
props: round/pine/palm trees, bush, stump, boulder, small rock, flowers,
mushroom, grass tuft, reeds, skull. Each run writes PNGs + a _*_sheet.png
contact sheet.
Status
- M1 — done: scaffold, palette, canvas primitives, pixelize ported.
- M2 — done: tileable terrain (7 biomes + water), 12 nature props, preview (contact sheets, 3x3 tiling check), procgen CLI.
- M3 — done: corner-based 16-tile autotile transitions between any two biomes, organic edges, dark rim, animated water foam. Seams exact by construction (bilinear corner field + edge-vanishing noise). Emits a bitmask manifest for the game renderer. 28 tests green.
- M4 — done: parametric buildings (3x3x3 house combos: size x roof x walls)
- 10-piece prop kit (fence/gate/well/barrel/crate/signpost/lantern/stall/ woodpile/haystack), gallery preview, village catalog. 35 tests green.
- M1-M4 polish pass — done: warm low-contrast terrain with soft details
(sprigs/pebbles/rare pale flowers) + new biomes (pine_floor, gravel, cobble,
dirt, path) and 6+ variants each to kill repetition; roads/paths as a linear
autotile feature (
roads.py, layouts cross/curve/plaza…) with grassy fringe; reference-level detailed houses (varied shingles, ridge cap, moss, 4-pane windows w/ muntins+shutters+glow, timber framing, stone base, framed door, capped chimney) + decor kit (bucket, sack, potted plant, flower box, lantern, axe-stump, …). Demo:python scripts/demo_scene.py. 40 tests green. - Anti-repetition pass — done: autotile tiles take a position-hashed
variantthat reseeds interior noise + fill (seams stay exact because noise vanishes at edges), so roads/maps no longer show a repeating pattern;terrain.ground_rect()bakes a large continuous ground field (soft patches flow across the whole area, no per-tile rhythm) for scene/chunk backgrounds; softer terrain details + more nature (flat rock, fallen branch, mushroom cluster). - M5+ — MCP server, export/integration, nngen (GPU). §11.
Autotile scheme
Corner 4-bit Wang / dual-grid: each tile's four corners (NW=1, NE=2, SE=4,
SW=8) mark upper vs lower terrain, giving 16 tiles. The boundary along any edge
is a pure function of that edge's two shared corners, so tiles that abut on the
dual grid are exactly seamless (0-mismatch, verified in tests). The renderer
picks a tile per world corner-cell from its 2x2 terrain values; the mapping is
written to <from>_<to>.manifest.json.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.