game-art-mcp
game-art-mcp
KI-gesteuertes Pixel-Art-Stilsystem und MCP-Server für die Art Direction von 2D-RPG-Spielen.
Zweck
Dieses Repository ist die maßgebliche Quelle für die Art Direction des Projekts. Jeder KI-Agent kann in dieses Repository wechseln, den Projektkontext über MCP abfragen und genau verstehen, was „unser Kunststil“ bedeutet — ohne auf den Gesprächsverlauf angewiesen zu sein.
Related MCP server: Aseprite MCP Server
Architektur
game-art-mcp/
├── project.yaml # Project config: which style is active
├── style/ # Version-controlled style definitions
│ └── fantasy_pixel_v1/ # Style v1 (YAML rules + style bible)
├── registry/ # Asset registry storage
│ ├── assets/ # One YAML file per registered asset
│ └── registry.yaml # Auto-generated index of all assets
├── memory/ # Art Memory storage (Phase 3)
│ ├── anchors/ # Style anchor YAML files
│ ├── references/ # Approved reference YAML files
│ ├── rejections/ # Rejection records
│ ├── decisions/ # Art decision records (ADR format)
│ ├── history.yaml # Style version evolution log
│ └── memory.yaml # Auto-generated memory index
├── src/
│ ├── style/ # Models, loader, validator
│ ├── assets/ # Asset registry (models + service)
│ │ ├── models/ # Zod schemas + TypeScript types
│ │ └── registry/ # AssetRegistry service (CRUD + query)
│ ├── memory/ # Art Memory (models, service, resolver)
│ │ ├── models/ # Zod schemas for anchors, references, rejections, decisions
│ │ ├── service/ # ArtMemoryService (CRUD + index)
│ │ └── resolver/ # ReferenceResolver (deterministic lookup)
│ ├── qa/ # Art QA engine (Phase 4)
│ │ ├── models/ # QA types, report schema, rule interface
│ │ ├── rules/ # 13 deterministic rules (7 categories)
│ │ ├── runner/ # QARunner orchestrator
│ │ └── history/ # QA history persistence
│ ├── providers/ # Provider Adapters (Phase 5)
│ │ ├── models/ # ProviderAdapter interface, types, error codes
│ │ ├── adapters/ # Adapter implementations (mock-provider)
│ │ ├── registry/ # ProviderRegistry (adapter lookup + capabilities)
│ │ ├── gateway/ # ProviderGateway (dispatch + artifact storage)
│ │ └── artifacts/ # ArtifactStore (immutable provenance)
│ ├── production/ # Production Orchestrator (Phase 6)
│ │ ├── models/ # Types, state machine, error codes
│ │ ├── orchestrator/ # ProductionOrchestrator (coordinator)
│ │ └── store/ # ProductionStore (YAML manifest persistence)
│ ├── versioning/ # Versioning & Approval (Phase 7)
│ │ ├── models/ # Types, lifecycle states, error codes
│ │ └── services/ # VersioningService (approval, versioning, promotion, audit)
│ ├── context/ # ArtContextService
│ └── mcp/ # MCP server + tools
│ └── tools/ # art-tools.ts, asset-tools.ts, memory-tools.ts, qa-tools.ts, provider-tools.ts, production-tools.ts, versioning-tools.ts
├── tests/ # Unit + integration tests
└── docs/ # Architecture, style system, phasesSchnellstart
npm install
npm run build
npm testMCP-Server ausführen
npm start
# or with custom root:
ART_MCP_ROOT=/path/to/project npm startStil validieren
npm run validateMCP-Werkzeuge
Stil-Werkzeuge (schreibgeschützt)
Werkzeug | Beschreibung |
| Vollständiger Art-Kontext (Projekt + Stil + alle Regeln) |
| Aktive Stildefinition |
| Bestimmte Regelkategorie (pixel_language, Sprachen, usw.) |
| Farbpalette mit semantischen Rollen |
| Stilkonfiguration validieren |
Asset-Werkzeuge (Lesen + Schreiben)
Werkzeug | Beschreibung |
| Asset anhand der ID abrufen |
| Assets suchen/filtern (Typ, Kategorie, Status, Tags) |
| Prüfen, ob eine Asset-ID registriert ist |
| Neues Asset mit vollständiger Validierung registrieren |
| Vorhandenes Asset aktualisieren (Teilupdate) |
| Asset als veraltet markieren |
| Asset archivieren |
| Registry-Index aus den Asset-Dateien neu aufbauen |
Memory-Werkzeuge (Lesen + Schreiben)
Werkzeug | Beschreibung |
| Memory-Übersicht: Anker, Entscheidungen, Ablehnungen, Referenzanzahl |
| Vollständige Stil-Erklärung mit Regeln, Ankern, Entscheidungen und Vermeidungen |
| Deterministische Referenzauflösung für einen bestimmten Kontext |
| Stilanker anhand der ID abrufen |
| Anker suchen (Kategorie, Status, Dimensionsfilter) |
| Neuen Stilanker hinzufügen |
| Freigegebene Referenz anhand der ID abrufen |
| Referenzen suchen (Rolle, Status, asset_id-Filter) |
| Neue freigegebene Referenz hinzufügen |
| Ablehnungsdatensatz anhand der ID abrufen |
| Ablehnungen suchen (Typ, Status, Grund-Filter) |
| Neuen Ablehnungsdatensatz hinzufügen |
| Künstlerische Entscheidung anhand der ID abrufen |
| Entscheidungen suchen (Statusfilter) |
| Neue künstlerische Entscheidung hinzufügen |
| Vollständige Historie der Stil-Entwicklung abrufen |
QA-Werkzeuge (schreibgeschützt)
Werkzeug | Beschreibung |
| QA-Prüfungen für ein einzelnes Asset ausführen (vollständiger Bericht) |
| QA-Prüfungen für mehrere Assets ausführen (Batch-Bericht) |
| QA-Gate-Prüfung — Ergebnis „bestanden/nicht bestanden“ für Freigabeworkflows |
| Alle verfügbaren QA-Regeln mit Definitionen auflisten |
| Vollständige Definition einer bestimmten QA-Regel anhand der ID abrufen |
| Erklären, warum eine bestimmte Regel für ein Asset fehlgeschlagen ist |
| QA-Ausführungsverlauf abrufen, optional nach Asset-ID gefiltert |
Provider-Werkzeuge (Lesen + Schreiben)
Werkzeug | Beschreibung |
| Alle registrierten Provider mit Metadaten auflisten |
| Detaillierte Metadaten für einen bestimmten Provider abrufen |
| Provider-Fähigkeiten abrufen (Operationen, Formate, Limits) |
| Health-Status des Providers prüfen |
| Kunstgenerierungsoperation erneut über einen Provider ausführen |
| Laufende Provider-Operation abbrechen |
| Status einer Operation anhand der ID abrufen |
| Artefakt-Details und Provenienz anhand der ID abrufen |
Produktions-Werkzeuge (Lesen + Schreiben)
Werkzeug | Beschreibung |
| Produktionsplan erstellen (Vorschau vor der Ausführung) |
| Produktions-Job erstellen (Plan + dauerhaft speichern, nicht starten) |
| Ausführung eines Produktions-Jobs starten |
| Aktuellen Jobstatus abrufen (Zusammenfassung) |
| Vollständige Jobdetails abrufen (Ereignisse, Versuche, Plan) |
| Fehlgeschlagenen Job wiederaufnehmen |
| Laufenden Job abbrechen |
| Versuchsverlauf für einen Job abrufen |
| Job genehmigen, der auf Freigabe wartet |
| Alle Produktions-Job-IDs auflisten |
Versionsverwaltungs-Werkzeuge (Lesen + Schreiben)
Werkzeug | Beschreibung |
| Kanonische (aktuelle) Version eines Assets abrufen |
| Details einer bestimmten Asset-Version abrufen |
| Vollständige Versionshistorie eines Assets abrufen |
| Zwei Versionen desselben Assets vergleichen |
| Versionsprovenienz inklusive Freigabedatensatz abrufen |
| Freigabe für ein Kandidaten-Asset beantragen |
| Freigabedatensatz anhand der ID abrufen |
| Kandidaten-Asset genehmigen |
| Kandidaten-Asset ablehnen |
| Änderungen an einem Kandidaten-Asset anfordern |
| Genehmigtes Kandidaten-Asset zur kanonischen Version befördern |
| Kanonische Version auf eine frühere Version zurücksetzen |
| Kanonisches Asset archivieren |
Stil- und QA-Werkzeuge sind schreibgeschützt. Asset-, Memory-, Provider-, Produktions- und Versionierungs-Werkzeuge unterstützen Lese- und Schreibzugriffe.
Asset-Registry
Das Asset-Registry (Phase 2) erfasst jedes Art-Asset im Projekt mit strukturierten Metadaten. Assets werden als einzelne YAML-Dateien unter registry/assets/ gespeichert und in registry/registry.yaml indiziert.
Wichtige Merkmale:
Semantische IDs — durch Punkte getrennt, klein geschrieben (z. B.
character.goblin.001)Stil-Verknüpfung — jedes Asset referenziert eine Stil-ID + Version
Beziehungen —
variant_of,derived_from,animation_ofusw.Statusverfolgung — draft, approved, rejected, deprecated, archived
Vollständige Validierung — Schema, Stilreferenz, Existenz der Quelldatei, Beziehungen
Siehe docs/ASSET-REGISTRY.md für die vollständige Dokumentation und docs/ASSET-METADATA.md für das Metadaten-Schema.
Art-Memory
Das Art-Memory-System (Phase 3) verleiht dem Repository dauerhaftes visuelles Wissen. Es speichert, was genehmigt wurde, was abgelehnt wurde und warum — sodass Agenten keine Gesprächshistorie benötigen, um die Art Direction des Projekts zu verstehen.
Wichtige Konzepte:
Stilanker — kanonische visuelle Beispiele, die den Stil definieren (siehe docs/STYLE-ANCHORS.md)
Genehmigte Referenzen — vertrauenswürdige Assets mit Rollen-Preferenzen. Ablehnungen — was NICHT passt, mit einem kontrollierten Vokabular an Gründen
Art-Entscheidungen — Datensätze im ADR-Format für visuelle Richtungsentscheidungen (siehe docs/ART-DECISIONS.md)
Referenz-Resolver — deterministische Referenzauflösung, die für jede Erstellungsaufgabe relevanten Kontext zurückgibt
Siehe docs/ART-MEMORY.md für die vollständige Dokumentation.
Art-QA
Das Art-QA-System (Phase 4) bietet deterministische, reproduzierbare Qualitätsprüfungen für Pixel-Art-Assets. Jeder Check ist regelbasiert mit Soll-/Ist-Werten und strukturierten Korrekturmaßnahmen — keine Bilderkennung, keine Embeddings, keine automatische Reparatur.
Wichtige Konzepte:
13 Regeln in 7 Kategorien (Technik, Dimensionen, Palette, Alpha, Pixel, Stil, Memuery)
3 Profile — „strict“ (schlägt bei Warnung fehl), „default“ (schlägt bei Fehler fehl), „lenient“ (schlägt nur bei kritischen Fehlern fehl)
Maschinenlesbare Berichte — JSON mit Ergebnissen pro Regel, Schweregrad und Korrekturmaßnahmen
Stil-Integration — liest Canvas-Größen, Farbgrenzen und Pixel-Regeln aus dem aktiven Stil
Memory-Integration — prüft total zurückgewiesene Richtungen und akzeptierte Art-Entscheidungen
QA-Gate — Ergebnis „bestanden/nicht bestandefor for CI und Freigabeworkflows
QA-Verlauf — persistentes Log aller Ausführungen pro Asset
Siehe docs/ART-QA.md für die vollständige Dokumentation.
Provider-Adapter
Das Provider-Adapter-System (Phase 5) fügt eine providerneutrale, anbieterunabhängige Schnittstelle zu externen Art-Generierungswerkzeugen hinzu. Anfragen laufen über ein Gateway, das Operationen validiert, an registrierte Adapter delegiert und generierte Artefakte mit unveränderlicher Provenienz speichert.
Wichtige Konzepte:
ProviderAdapter-Schnittstelle — Metadaten, Fähigkeiten, Health, Ausführung, Abbruch
Artefakte — Rohe Provider-Ausgabe mit unveränderlicher Provenienz (noch keine Assets)
Capabilities — Details pro Operation (Formate, maximale Auflösung)
Dry-run — Anfragen validieren, ohne Ausgabe zu generieren
Mock-Provider — eingebauter Testadapter mit Fehler-/Timeout-Modi
Keine automatische Auswahl — Agenten müssen einen Provider eExplicitly wählen
Siehe docs/PROVIDERS.md für die vollständige Dokumentation.
Produktions-Orchestrator
Der Produktions-Orchestrator (Phase 6) koordiniert den gesamten Lebenszyklus der Art-Asset-Generierung: request validation, Stil-/Referenz-/Provider-Auflösung, Ausführung, QA, Wiederholung und Freigabeprüfung.
Wichtige Konzepte:
Koordinator, keine Quelle der Wahrheit — delegiert an Stil, Qualitätssicherung, Anbieter und Registry
Zustandsmaschine — 9 Status mit validierten Übergängen (erstellt bis abgeschlossen/fehlgeschlagen/abgebrochen)
11 Produktionsstufen — REQUEST_VALIDATION bis APPROVAL_GATE
Begrenzte Wiederholung — konfigurierbare max_attempts (Standard 3) mit Reparaturplänen bei QA-Fehlern
Genehmigungsgrenze — stoppt bei
awaiting_approval, genehmigt niemals automatischPlan-Veraltung — erkennt Stilversionsabweichungen vor der Ausführung
YAML-Persistenz — ein manifest.yaml pro Auftrag in
production/<job_id>/Ereignishistorie — nur-anhängendes Protokoll aller Zustandsänderungen pro Auftrag
Siehe docs/PRODUCTION.md für die vollständige Dokumentation.
Versionierung & Genehmigung
Das Versionierungs- & Genehmigungssystem (Phase 7) fügt unveränderliche Asset-Versionierung, explizite Genehmigungsworkflows und eine vollständige Prüfpfad hinzu. Keine Version wird jemals gelöscht; kein Asset wird jemals automatisch genehmigt.
Wichtige Konzepte:
Asset-Lebenszyklus — 8 Zustände: Entwurf, ausstehende_Genehmigung, genehmigt, abgelehnt, Änderungen_angefordert, befördert, ersetzt, archiviert
Genehmigungsworkflow — anfordern, genehmigen, ablehnen, Änderungen_anfordern mit strukturiertem Feedback
Genehmigungsrichtlinie — konfigurierbar:
requires_qa_pass,allow_agent_approval,requires_humanUnveränderliche Versionen — monotone Inkrementierung, Elternverfolgung, vollständige Herkunft pro Version
Kanonischer Zeiger — verfolgt, welche Version aktuell ist; wird bei Beförderung/Rollback aktualisiert
Beförderung — Vergleich-und-Tausch mit QA-Gate und Genehmigungs-Gate
Rollback — setzt den kanonischen Zeiger auf eine frühere Version zurück, löscht niemals die Historie
Prüfprotokoll — 9 Ereignistypen, nur-anhängend, unveränderlich
Akteursidentität — Mensch, Agent, System, Anbieter bei jedem Datensatz verfolgt
Siehe docs/VERSIONING.md für die vollständige Dokumentation.
Aktuelle Phase
Phase 7 — Versionierung & Genehmigung (abgeschlossen)
Siehe docs/PHASES.md für die vollständige Roadmap.
Stilsystem
Stile sind strukturierte YAML-Dateien, die maschinenlesbare künstlerische Richtung darstellen:
style.yaml— Identität, Leinwandgrößen, Skalierungpalette.yaml— Farben mit semantischen Rollenpixel-rules.yaml— Pixel-Art-Einschränkungenoutline.yaml— Konturregelnshape-language.yaml— visuelle Sprachelighting.yaml— Lichtrichtung und -regelnanimation.yaml— Bildanzahl, FPS, Einschränkungen
Siehe docs/STYLE-SYSTEM.md für Details.
Available Tools
126 toolsart.aseprite.exportB
Export a canvas session to Aseprite format. Saves the canvas as PNG and generates the Aseprite CLI command to convert it. Commands are returned for external execution.
Inputs: session_id (required), output_name.
art.aseprite.layersA
Generate a Lua script and Aseprite command to set up standard layer structure. Writes the script to disk and returns the command to execute it.
Inputs: session_id (required), layers.
art.aseprite.spritesheetB
Generate a spritesheet export command for a canvas session. The canvas PNG is saved and a CLI command is returned to produce a packed spritesheet via Aseprite.
Inputs: session_id (required), columns, padding.
art.aseprite.validateB
Generate a validation Lua script and Aseprite command to check sprite constraints (palette size, dimensions). Writes the script to disk and returns the command.
Inputs: width (required), height (required), max_colors (required), session_id (required).
art.asset.approval.inspectB
Get an approval record by ID.
Inputs: approval_id (required).
art.asset.approval.requestC
Request approval for a candidate asset.
Inputs: actor (required), style (required), asset_id (required), candidate_id (required), +3 more.
art.asset.approveC
Approve a candidate asset.
Inputs: actor (required), approval_id (required), reason.
art.asset.archiveB
Archive an asset (remove from active use).
Inputs: id (required).
art.asset.archive_versionC
Archive a canonical asset, removing it from active use.
Inputs: actor (required), asset_id (required).
art.asset.compareB
Compare two versions of an asset.
Inputs: asset_id (required), version_a (required), version_b (required).
art.asset.currentB
Get the canonical (current) version number of an asset.
Inputs: asset_id (required).
art.asset.deprecateC
Mark an asset as deprecated.
Inputs: id (required).
art.asset.existsA
Check whether an asset with the given ID exists in the registry.
Inputs: id (required).
art.asset.findA
Search and filter registered assets. All parameters are optional; omit all to return every asset.
Inputs: tags, type, status, category, +2 more.
art.asset.getA
Get a registered asset by its ID. Returns the full asset record.
Inputs: id (required).
art.asset.historyC
Get the full version history of an asset.
Inputs: asset_id (required).
art.asset.inspect_versionB
Get details of a specific asset version.
Inputs: version (required), asset_id (required).
art.asset.promoteC
Promote an approved candidate to become the canonical version.
Inputs: actor (required), style (required), approval_id (required), artifact_id (required), +7 more.
art.asset.provenanceC
Get version provenance including approval record.
Inputs: version (required), asset_id (required).
art.asset.rebuild_indexA
Rebuild the asset registry index from all registered asset files.
Inputs: no named parameters.
art.asset.registerA
Register a new asset in the registry. Validates schema, style reference, source file, and relationships.
Inputs: id (required), name (required), type (required), style (required), +8 more.
art.asset.rejectC
Reject a candidate asset.
Inputs: actor (required), approval_id (required), reason, feedback.
art.asset.request_changesC
Request changes on a candidate asset.
Inputs: actor (required), feedback (required), approval_id (required), reason.
art.asset.rollbackB
Rollback the canonical version to a previous version.
Inputs: actor (required), asset_id (required), target_version (required), expected_current_version (required).
art.asset.updateC
Update an existing asset. Provide the asset ID and a patch object with fields to change.
Inputs: id (required), patch (required).
art.canvas.analyze_refA
Analyze a reference image: extract dimensions, color palette with counts, and 2D color map.
Inputs: image_path (required).
art.canvas.batch_drawA
Execute multiple draw commands in a single call for efficiency. Each command references a canvas tool name and its params.
Inputs: commands (required), session_id (required).
art.canvas.clearB
Clear the entire canvas, optionally filling with a color.
Inputs: session_id (required), color.
art.canvas.createB
Create a new pixel canvas session for drawing.
Inputs: session_id (required), width, height, asset_id, +1 more.
art.canvas.draw_circleB
Draw a circle on the canvas.
Inputs: cx (required), cy (required), color (required), radius (required), +2 more.
art.canvas.draw_ditherB
Apply a dither pattern to a rectangular region.
Inputs: x (required), y (required), width (required), color1 (required), +4 more.
art.canvas.draw_lineA
Draw a line between two points using Bresenham algorithm.
Inputs: x0 (required), x1 (required), y0 (required), y1 (required), +2 more.
art.canvas.draw_noiseC
Fill a region with random noise using specified colors.
Inputs: x (required), y (required), width (required), colors (required), +3 more.
art.canvas.draw_rectB
Draw a rectangle on the canvas.
Inputs: x (required), y (required), color (required), width (required), +3 more.
art.canvas.fillB
Flood fill an area starting from a point.
Inputs: x (required), y (required), color (required), session_id (required).
art.canvas.get_pixelB
Read the color of a single pixel on the canvas.
Inputs: x (required), y (required), session_id (required).
art.canvas.get_stateA
Get the current canvas state as a color map (2D hex array) or base64 PNG.
Inputs: session_id (required), format.
art.canvas.saveA
Save the current canvas to a PNG file.
Inputs: session_id (required), output_path.
art.canvas.set_pixelB
Set a single pixel on the canvas.
Inputs: x (required), y (required), color (required), session_id (required).
art.critic.evaluateB
Run the art critic to evaluate canvas quality with spatial scoring and issue detection.
Inputs: session_id (required), profile.
art.critic.refineC
Execute refinement actions from a critique to improve canvas quality.
Inputs: actions (required), session_id (required).
art.detail.cavityB
Add cavity details (dark center with light rim) to a region.
Inputs: x (required), y (required), seed (required), count (required), +5 more.
art.detail.clean_clustersC
Cleanup pass: merge orphan/isolated pixels into nearest larger cluster.
Inputs: session_id (required), min_cluster_size.
art.detail.clusterC
Create a pixel cluster on the canvas at a given position with a specified shape.
Inputs: cx (required), cy (required), size (required), color (required), +3 more.
art.detail.crackB
Add crack patterns to a region (stone, ice surfaces).
Inputs: x (required), y (required), seed (required), color (required), +4 more.
art.detail.edge_breakB
Apply edge breakup to avoid perfectly geometric silhouettes.
Inputs: seed (required), amount (required), session_id (required), preserve_outline.
art.detail.executeC
Run the full detail pass pipeline on an existing canvas session. Requires content already drawn on canvas.
Inputs: materials (required), session_id (required), seed, profile.
art.detail.grainA
Add directional grain lines to a region (wood, cloth).
Inputs: x (required), y (required), seed (required), color (required), +5 more.
art.detail.highlightC
Add highlight clusters to a region for specular/light reflection.
Inputs: x (required), y (required), seed (required), color (required), +4 more.
art.detail.materialB
Apply a full material treatment (all texture patterns for that material) to a region.
Inputs: x (required), y (required), seed (required), width (required), +3 more.
art.detail.passC
Run a single named detail pass on a canvas session.
Inputs: materials (required), pass_name (required), session_id (required).
art.detail.planB
Create a detail art plan for an asset, specifying subject, resolution, materials, and detail profile.
Inputs: width (required), height (required), subject (required), materials (required), +2 more.
art.detail.shadowA
Add shadow clusters to a region for depth/occlusion.
Inputs: x (required), y (required), seed (required), color (required), +4 more.
art.detail.speckleA
Add scattered speckle dots to a region (dirt, stone, skin).
Inputs: x (required), y (required), seed (required), width (required), +4 more.
art.detail.textureC
Apply a texture pattern to a rectangular region on the canvas.
Inputs: x (required), y (required), seed (required), width (required), +5 more.
art.get_paletteB
Returns the color palette for the active style.
Inputs: no named parameters.
art.get_project_contextA
Returns the full art context: project config, active style definition, all rule categories, and palette.
Inputs: no named parameters.
art.get_styleA
Returns the active style definition. Optionally specify style_id and version to look up a specific style.
Inputs: version, style_id.
art.get_style_rulesB
Returns the rules for a specific category of the active style.
Inputs: category (required).
art.memory.add_anchorC
Add a new style anchor to the art memory.
Inputs: id (required), name (required), style (required), assets (required), +6 more.
art.memory.add_decisionC
Add a new art decision record to the art memory.
Inputs: id (required), style (required), title (required), reason (required), +6 more.
art.memory.add_referenceC
Add a new approved reference to the art memory.
Inputs: id (required), roles (required), style (required), status (required), +5 more.
art.memory.add_rejectionB
Add a new rejection record to the art memory.
Inputs: id (required), type (required), style (required), status (required), +5 more.
art.memory.explain_styleA
Explain the current or specified style: core rules, canonical anchors, important decisions, things to avoid, and history.
Inputs: style_id, style_version.
art.memory.find_anchorsC
Search for style anchors with optional filters.
Inputs: status, category, style_id, dimension, +1 more.
art.memory.find_decisionsB
Search for art decision records with optional filters.
Inputs: status, style_id, style_version.
art.memory.find_referencesC
Search for approved references with optional filters.
Inputs: role, status, asset_id, style_id, +1 more.
art.memory.find_rejectionsC
Search for rejection records with optional filters.
Inputs: type, reason, status, style_id, +1 more.
art.memory.get_anchorA
Get a style anchor by its ID.
Inputs: id (required).
art.memory.get_decisionA
Get an art decision record by its ID.
Inputs: id (required).
art.memory.get_referenceC
Get an approved reference by its ID.
Inputs: id (required).
art.memory.get_rejectionA
Get a rejection record by its ID.
Inputs: id (required).
art.memory.get_style_historyB
Get the full style evolution history.
Inputs: no named parameters.
art.memory.get_summaryA
Get a summary of the art memory: anchor count, active decisions, active rejections, total references.
Inputs: no named parameters.
art.memory.insightsA
Get style insights derived from approved assets — patterns, preferences, and learned conventions from the asset corpus.
Inputs: no named parameters.
art.memory.negativeA
Get negative patterns (rejected examples) for a type/material — query rejections filtered by type hints in reasons.
Inputs: type, material.
art.memory.resolve_referencesD
Resolve relevant references, anchors, decisions, and rejections for a given context. Returns categorized results.
Inputs: type, category, style_id, dimensions, +1 more.
art.memory.signatureA
Compute a visual signature (deterministic fingerprint) for a canvas session — color distribution, cluster stats, edge complexity, etc.
Inputs: session_id (required).
art.pipeline.analyzeB
Analyze a reference image. Returns palette, silhouette, composition, pixel characteristics. Provide either a file path or reference_id of an already-uploaded reference.
Inputs: pose, facing, subject, category, +2 more.
art.pipeline.planA
Build a 7-stage ArtPlan (silhouette-first) from an ArtSpecification. The plan defines ordered generation stages with QA checkpoints.
Inputs: reference_id (required), max_iterations.
art.pipeline.scoreC
Compute visual quality score for a pixel art asset. Scores silhouette, palette, composition, pixel cleanliness, and style match.
Inputs: asset_path (required), reference_id.
art.pipeline.specB
Build an ArtSpecification from a reference analysis + active style. The spec defines canvas, palette, outline, shading constraints for pixel art generation.
Inputs: reference_id (required), subject, max_colors, canvas_width, +1 more.
art.production.approveB
Approve a production job that is awaiting approval, moving it to completed.
Inputs: job_id (required).
art.production.attemptsB
Get the attempt history for a production job.
Inputs: job_id (required).
art.production.cancelC
Cancel a production job.
Inputs: job_id (required).
art.production.createB
Create a production job (plan + persist, does not start execution).
Inputs: output (required), subject (required), asset_type (required), request_id (required), +5 more.
art.production.inspectC
Get full details of a production job including events, attempts, and plan.
Inputs: job_id (required).
art.production.listB
List all production job IDs.
Inputs: no named parameters.
art.production.planB
Create a production plan for an art asset (inspect before executing).
Inputs: output (required), subject (required), asset_type (required), request_id (required), +5 more.
art.production.resumeA
Resume a failed or interrupted production job.
Inputs: job_id (required).
art.production.startC
Start executing a production job.
Inputs: job_id (required).
art.production.statusC
Get the current status of a production job.
Inputs: job_id (required).
art.provider.analyzeB
Analyze a PNG file using the canvas-analysis provider. Runs local analysis (signature, clusters, palette) without external APIs.
Inputs: operation (required), source_path (required), style, asset_id, +2 more.
art.provider.artifactB
Get details of a generated artifact.
Inputs: artifact_id (required).
art.provider.cancelC
Cancel a running provider operation.
Inputs: operation_id (required).
art.provider.capabilitiesB
Get the capabilities of a specific provider.
Inputs: provider_id (required).
art.provider.executeC
Execute an art generation operation via a provider.
Inputs: style (required), operation (required), provider_id (required), output, +6 more.
art.provider.fallbackC
Execute an art operation with fallback — tries the primary provider, falls back to a secondary on failure.
Inputs: style (required), operation (required), provider_id (required), output, +7 more.
art.provider.getA
Get detailed metadata for a specific provider.
Inputs: provider_id (required).
art.provider.healthB
Check the health status of a specific provider.
Inputs: provider_id (required).
art.provider.health_allA
Check health of all registered providers. Returns a map of provider ID to health status with latency and consecutive failure count.
Inputs: no named parameters.
art.provider.listA
List all registered art providers with their metadata.
Inputs: no named parameters.
art.provider.operationC
Get the status of a provider operation.
Inputs: operation_id (required).
art.qa.assetA
Run QA checks on a single asset. Returns a full QA report with all rule results.
Inputs: asset_id (required), profile.
art.qa.batchA
Run QA checks on multiple assets at once. Returns a batch report with individual results.
Inputs: asset_ids (required), profile.
art.qa.explain_failureA
Explain why a specific QA rule failed for an asset. Returns rule definition, issue details, and context.
Inputs: rule_id (required), asset_id (required).
art.qa.gateB
Run a QA gate check on an asset. Returns pass/fail verdict with full report.
Inputs: asset_id (required), profile.
art.qa.historyC
Get QA run history. Optionally filter by asset ID.
Inputs: asset_id.
art.qa.list_rulesA
List all available QA rules with their definitions.
Inputs: no named parameters.
art.qa.ruleA
Get the full definition of a specific QA rule by its ID.
Inputs: rule_id (required).
art.ref.canonicalA
Get canonical assets for a type/material — filters retrieve results to only canonical status entries.
Inputs: type, material.
art.ref.packB
Build a reference pack for generation — combines style constraints, canonical references, negative examples, and material palette.
Inputs: type, category, material, max_negative, +1 more.
art.ref.retrieveB
Smart ranked reference retrieval — scores and returns top assets by canonical status, style match, material match, and recency.
Inputs: type, category, material, max_results.
art.style.bootstrapA
Get session bootstrap context — assembles project config, style, canonical assets, anti-patterns, and recent decisions into one payload.
Inputs: no named parameters.
art.style.checkA
Run style consistency check on a canvas session — compares canvas content against compiled style constraints.
Inputs: session_id (required), material.
art.style.compileB
Compile style constraints for a subject — merges style definition, optional material profile, and resolution into a flat StyleConstraints object.
Inputs: width, height, profile, material.
art.style.repairA
Generate a repair plan from a style consistency check — identifies actions to fix style issues on a canvas.
Inputs: session_id (required), material.
art.tilemap.build_opsB
Generate TiledMCP operations from a MapSpecification. Returns the operation sequence for TiledMCP provider execution.
Inputs: map_id (required).
art.tilemap.create_specC
Create a MapSpecification for tilemap generation. Defines map size, biome, layers, objects, terrain rules.
Inputs: biome (required), width (required), height (required), layers (required), +6 more.
art.tilemap.list_tilesetsA
List all registered tileset definitions.
Inputs: no named parameters.
art.tilemap.register_tilesetB
Register a tileset definition with semantic tile metadata. Each tile gets an ID, terrain, walkability, collision info.
Inputs: tiles (required), asset_id (required), tile_size (required), tileset_id (required), +2 more.
art.tilemap.resolve_tilesA
Resolve semantic tile names to tileset local IDs. Use to verify tile names before building a map.
Inputs: intents (required), tileset_id (required).
art.tilemap.validateA
Validate a MapSpecification against its tileset. Checks missing tiles, invalid refs, style mismatch, object bounds.
Inputs: map_id (required).
art.validate_styleB
Validates the active style configuration and returns any errors found.
Inputs: no named parameters.
art.workflow.guideA
Generate a step-by-step painting guide for the AI to follow when drawing a subject. Enforces strict palette discipline, form-first/detail-last hierarchy, and per-step color limits. The guide is TEXT instructions — read it and follow each step sequentially.
Inputs: subject (required), width, height, profile, +1 more.
art.workflow.paletteA
Get the strict palette for a material — returns role-to-hex mappings, maximum color count, and forbidden colors. Use this before painting to know EXACTLY which colors are allowed.
Inputs: material.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 126 tools
The hierarchical namespaces help, but the sheer number of tools creates real overlap: art.provider.analyze vs art.pipeline.analyze, art.asset.archive vs art.asset.archive_version, and art.qa.asset vs art.critic.evaluate vs art.style.check all have fuzzy boundaries. An agent could easily pick the wrong one without very careful descriptions.
Most tools follow a predictable art.<category>.<action> pattern with snake_case verbs, which is highly readable. However, a few top-level tools like art.validate_style, art.get_style, and art.get_palette break the convention, and verbs are mixed across add/register/create for similar operations.
126 tools is an extreme count for any MCP server, far beyond the 25+ threshold. Many micro-tools—individual detail effects, memory CRUD variants, and Aseprite command generators—could be consolidated into broader operations, making the surface unnecessarily hard to navigate.
The tool surface is remarkably comprehensive for a game-art domain: asset lifecycle, versioning, approvals, style memory, QA, providers, canvas painting, production jobs, tilemaps, detail passes, Aseprite export, and workflow guidance are all covered. Major workflows have no obvious dead ends.
Maintenance
Related MCP Connectors
AI game assets for agents: consistent sprites, 2D animations, tiles, maps, music and engine exports.
Generate, edit, style-match, turn, and animate game-ready pixel art at exact sizes.
Pixel art for your game, from your coding agent. Generate sprites, items and tiles in your game's style (64 to 1024 px), make variants, reskins and N/E/S/W directions, then edit everything in a full pixel art editor that syncs into your project. Works with Claude, Claude Code, Cursor and ChatGPT; sign in with OAuth, no API key.
Build a game's 2D art layer with your agent: characters, animations, tilesets, levels, 5 engines.
Related MCP Servers
- AlicenseNot gradedqualityCmaintenanceEnables AI coding agents to create, draw, and refine Aseprite documents through deterministic canvas, layer, frame, tag, and palette tools, with pixel analyzers, style validation, and preview/diff visual feedback. It also supports spritesheet and animated GIF export, producing multi-layer game-ready assets.1MIT
- AlicenseBqualityCmaintenanceEnables users to automate the creation, editing, animation, and export of pixel-art game assets through a multi-agent pipeline that derives style specs and blueprints, generates Lua build scripts, and drives the canvas editor to produce .aseprite files, spritesheets, atlases, GIFs, and previews.123MIT
- AlicenseNot gradedqualityCmaintenanceEnables AI agents to create and edit pixel-art projects — sprites, props, palettes, eight-direction animation, and tile-based environments — through a stdio MCP server with discovery, help, and call tools. Agents can build from silhouette to detail, inspect native-scale renders, silhouettes, value maps, and GIF playback, then save every change as versionable source.MIT
- AlicenseBqualityCmaintenanceEnables MCP-capable coding agents to act as a deterministic technical art director and multi-engine pipeline builder, locking AI-generated 2D/2.5D game assets to consistent camera angles, palettes, rigs, atlases, and exports across Blender, ComfyUI, UniRig, LibreSprite, Godot 4, Unity, and Unreal Engine 5.24MIT