Game Artist MCP
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Game Artist MCPCreate a 64x64 fantasy potion icon and render a PNG preview."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Game Artist MCP
An AI-native deterministic game-art workstation for agent hosts.
Game Artist is not another general MCP control plane and does not use a generative image model as its core art engine.
Its job is to give an LLM explicit game-art operations:
LLM
↓
Game Artist MCP
↓
Art IR / explicit geometry
↓
render
↓
visual inspection
↓
targeted edit
↓
game-ready assetRelationship to FolderForge
FolderForge is the preferred production host/control plane:
ChatGPT / MCP host
↓
FolderForge
↓
Game Artist
↓
deterministic art coreFolderForge owns generic infrastructure such as transport, authentication, policy, audit, sandboxing, plugin lifecycle and broad Godot control.
Game Artist owns art-domain capabilities such as vector/cutout representation, pivots, animation recipes, style constraints and art QA.
The core remains usable directly through MCP without FolderForge.
See:
Related MCP server: frameforge-mcp
Current state
POC-001/002: deterministic vector art engine and 11-tool stdio MCP with FolderForge integration.
POC-003: render → image inspection → targeted edit loop (mechanics only).
POC-004: Spirit Lantern cutout/Godot exporter and headless runtime verification.
POC-005A: ArtStyleProfile, safe structural evidence auditor, and three agent-directed local MCP trials (cultivation sword, medicinal pouch, stone censer) under benchmarks/poc005a/runs. Each preserves two actual PNG renders and a correction on the same asset. Their local structural audit is included in
npm test.
Evidence boundary: The three live sessions were executed through a local stdio MCP transport, not an independently audited provider-origin agent host. Their image critique is SELF_REVIEW. The repository therefore retains AGENT_PROVENANCE_PENDING and does not claim artistic FULL_PASS, independent aesthetics approval or production art quality. See POC-005A benchmark protocol.
POC-005B is merged: opt-in material-aware Art IR, deterministic gradient/hand-directed surface marks/bounded glow, plus controlled same-geometry 64/128-pixel A/B review bundles for those three earlier props. The review bundles are UNSCORED. A fourth original compass asset was constructed through a new self-selected local MCP session (not an evaluator-supplied held-out prompt): 28 nodes, 14 MCP calls, 3 rendered previews and 2 image-grounded corrections. The latest correction restored an open suspension loop that an inappropriate appearance fill made visually solid. Historical revision-2 media are preserved. Technical and style gates pass, but external authorship attestation and independent visual scoring remain pending. See POC-005B review protocol and latest handoff.
Hard limits: Material-aware Godot export is currently unsupported (UNSUPPORTED_GODOT_APPEARANCE), even though the legacy Spirit Lantern passes headless Godot. Green CI proves technical behavior, not aesthetic quality, autonomy or production art readiness.
Local development
Requirements:
Node.js 22+
rsvg-convertfrom librsvg for PNG previews
Run the raw MCP server:
node src/mcp/server.mjs --workspace ./.game-artistRun the direct protocol smoke:
npm run smoke:mcpTo verify the FolderForge integration before the compatible package is published, build the exact source revision used by CI:
git clone https://github.com/roronoazoroshao369/FolderForge.git ../FolderForge
git -C ../FolderForge checkout 59c1096167a81bef24a07fa88453f5c50e0fe64a
npm --prefix ../FolderForge ci --ignore-scripts
npm --prefix ../FolderForge run build
node ../FolderForge/dist/main.js plugin validate .
node ../FolderForge/dist/main.js plugin test . --call health --args-json '{}'Do not infer npm availability from FolderForge's source package.json; POC-002 pins source evidence until a compatible public package is actually published.
POC-005C1 (implemented technical workflow, still UNSCORED): Offline Hybrid Independent Visual Evaluation now reconstructs three actual baseline/enhanced props, checks geometry and 64/128px PNG bytes, creates reviewer-only offline A/B packages, accepts evaluator-verified blind human scoring, locks forms before mapping reveal and computes deterministic VISUAL_GO/ARTISTIC_NO_GO/REVIEW_PENDING. AI critic input is ADVISORY_ONLY, never a human reviewer. Run npm run poc:005c1:dry to generate an UNSCORED CI reviewer kit; its result must stay TECHNICAL_PASS, REVIEW_PENDING, and zero independent reviewers. Source: spec, TDD plan, reviewer instructions. The shipped implementation was checked by SELF_REVIEW because no separate subagent/code-review tool was exposed; do not mislabel that as independent code review.
POC-005C1 trust-boundary follow-up (PR #19): The offline reviewer must now explicitly attest HUMAN/NOT_ART_AUTHOR/NOT_EXPOSED; evaluator verification must bind to the same frozen session/package and acknowledgement timestamp, and a separate distinct out-of-band evidence reference is required for each reviewer. Previous reviewer kits without the declaration are incompatible: issue a NEW frozen review session; never manufacture or retroactively edit human attestations. Mechanical tests do not prove reviewer identity or artwork merit.
Next external evidence gate: obtain at least one real independent non-author human blind evaluation at 64px/128px (ideally two, frozen quorum), and independently obtain a provider-attested image-visible agent run on an evaluator-assigned new brief (separate POC-005C2). No human visual scores or provider-origin attestations have been supplied; the repository MUST retain VISUAL_REVIEW_PENDING / AGENT_PROVENANCE_PENDING until authentic evidence exists. A green CI or a synthetic fixture is not artistic GO or production-readiness proof.
This server cannot be deployed
Maintenance
Related MCP Connectors
Generate game-ready 3D models, textures, and audio from natural language, over MCP.
OCR, transcription, file extraction, and image generation for AI agents via MCP.
Brand-safe MCP for AI agents to create editable, on-brand graphics and automate variants.
Generate on-brand images from your AI agent: design, edit, and render templates over MCP.
Related MCP Servers
- AlicenseNot gradedqualityAmaintenanceA FastMCP server providing structured SVG authoring tools for LLMs, enabling create, edit, and render SVG graphics through hierarchical primitives, gradients, paths, and reusable resources.126 PyPI6GPL 2.0
- AlicenseNot gradedqualityBmaintenanceMCP server that renders AI-authored documents to images and returns typed diagnostics for self-correction. It also provides visual QA metrics, raster-to-vector reconstruction, and image-to-draft proposals.MIT
- FlicenseNot gradedqualityBmaintenanceExposes a deterministic 2D design canvas as MCP tools, letting an LLM client inspect canvas state, place and modify elements, track reward, render images, and reset the environment to build marketing assets from a natural language brief.-
- AlicenseNot gradedqualityBmaintenanceEnables MCP clients to create, inspect, edit, render, and export SVG/Inkscape documents through typed drawing tools, with a deterministic offline backend and an optional bridge to a locally running Inkscape instance for limited live editing and snapshots.4BSD 3-Clause