dcc-mcp-krita

Typed document, layer, painting, save, and export automation for Krita.

Illustrative workflow generated with OpenAI ImageGen from the retained source in docs/images/sources; it is not a Krita screenshot or host-validation artifact.

The adapter combines an external DCC-MCP service with a small Krita Python extension. The extension accepts authenticated loopback JSON-lines requests, queues every LibKis call onto Krita's UI main thread, and exposes only a fixed typed command catalog. It never evaluates caller-provided Python or action IDs.

Capabilities

• discover the live bridge and list or inspect open documents;
• create bounded RGBA/U8 documents or open supported files under allowed roots;
• inspect layer hierarchy and create, activate, rename, lock, hide, fade, or delete paint layers;
• write bounded typed-color rectangles directly to paint-layer pixels;
• preserve layered work as .kra and export PNG, JPEG, WebP, or TIFF artifacts;
• return artifact size and SHA-256 for downstream Blender, Godot, or engine import validation;
• flatten or discard modified documents only with explicit typed confirmation.

The bundled krita-document-authoring Skill exposes 16 tools. Freehand brush simulation, arbitrary Python, generic menu actions, and unrestricted filesystem access are intentionally outside the contract.

Install

pip install dcc-mcp-krita
dcc-mcp-krita-install

Restart Krita, open Settings → Configure Krita → Python Plugin Manager, and enable DCC MCP Krita. Configure allowed file roots before starting both Krita and the adapter:

$env:DCC_MCP_KRITA_ALLOWED_ROOTS = "C:\art\textures;D:\project\sprites"
krita.exe
dcc-mcp-krita

On POSIX systems, separate roots with : instead of ;. The bridge defaults to 127.0.0.1:3848; DCC_MCP_KRITA_BRIDGE_PORT changes it for both processes. The extension and adapter share a random token through the current user's ~/.dcc-mcp/krita-bridge-token; set DCC_MCP_KRITA_BRIDGE_TOKEN or DCC_MCP_KRITA_BRIDGE_TOKEN_FILE when deployment policy requires another secret source.

Inspect installation state without revealing the token:

dcc-mcp-krita-doctor

Safety model

• loopback-only authenticated transport with bounded requests and responses;
• a bounded main-thread queue; socket workers never call Krita APIs;
• workspace-root, suffix, file-size, pixel-count, layer-count, and timeout limits;
• instance-scoped document and layer IDs instead of caller-provided expressions;
• explicit overwrite, flatten, and discard opt-ins;
• batch-mode export with bounded format options and artifact hashing;
• staged extension installation with rollback if replacement fails.

See Architecture for the runtime and failure contracts.

Development

python -m pip install -e ".[dev]"
python -m pytest -q
python -m ruff check src tests tools bridge
python -m ruff format --check src tests tools bridge
python tools/lint_skills.py
python -m build
python -m twine check dist/*

The host implementation follows Krita's official Python plug-in guide and LibKis API.