Krita MCP Server & CLI

Let AI agents paint in Krita via the Model Context Protocol.

This subproject provides the core implementation of the Krita-MCP ecosystem, including the FastMCP server, the krita CLI, and the high-performance Python plugin.

🛠️ Components

1. Krita Plugin (krita-plugin/) — A Python plugin for Krita that exposes a thread-safe HTTP server on localhost using the configured port (default 5678).
2. MCP Server (src/krita_mcp/) — A FastMCP server exposing 40+ painting and manipulation tools.
3. Krita CLI (src/krita_cli/) — A Typer-based command line interface for human operators.
4. Krita Client (src/krita_client/) — A reusable, fully-typed Python library for Krita automation.

🚀 Setup

1. Krita Plugin Installation

Copy the contents of krita-plugin/ to your Krita resources:

• Windows: %APPDATA%\krita\pykrita\
• Linux: ~/.local/share/krita/pykrita/
• macOS: ~/Library/Application Support/krita/pykrita/

Enable "Krita MCP Bridge" in Krita (Configure Krita → Python Plugin Manager) and restart.

2. Environment Setup

# Install dependencies with uv
uv sync

3. Windows Validation Checklist

Before debugging the plugin, make sure the local Python runtime is healthy:

1. python -c "import ssl, ctypes" must succeed.
2. python scripts/windows_preflight.py should report a clean Windows setup.
3. uv run pytest tests/e2e/test_e2e_mock.py should pass without Krita.
4. The plugin files must exist under %APPDATA%\krita\pykrita\kritamcp\.
5. Krita must be restarted after enabling Krita MCP Bridge in the Plugin Manager.
6. The plugin should expose http://127.0.0.1:<configured-port>/health once Krita finishes loading.

If Krita starts but /health never appears, check the diagnostic artifacts in your home directory:

• ~/kritamcp_startup.log
• ~/kritamcp_diag.log

If those logs are missing entirely, the plugin likely was not enabled or did not load during startup.

🔧 CLI Commands

The krita CLI is grouped into logical subcommands:

• Painting: stroke, fill, clear, draw-shape
• Layers: layers list, layers create, layers select, layers delete
• Selection: selection select-rect, selection transform, selection save-channel
• Canvas: canvas-info, current-color, current-brush
• Session: history, replay, rollback, batch
• System: health, config, capabilities, security-status

Run uv run krita --help for full details.

🤖 MCP Server Tools (40 Total)

The MCP server exposes a vast range of capabilities to AI agents:

⚡ Performance

The plugin uses numpy-accelerated direct pixel manipulation for rendering. This ensures that strokes and shapes are rendered significantly faster than standard Python loops, especially on high-resolution canvases.

🔒 Security

Krita MCP includes a built-in security layer to protect your system:

• Path Sanitization: Restricts file operations to allowed directories.
• Resource Limits: Prevents OOM by limiting max canvas dimensions and layer counts.
• Rate Limiting: Throttles rapid command execution.
• Security Tool: krita_security_status allows agents to check active limits.

📄 License

MIT