Krita MCP - Paint with Claude

An MCP server that lets Claude paint in Krita. Paint, see the result, adjust, repeat.

How It Works

1. A Krita plugin runs an HTTP server inside Krita
2. An MCP server bridges Claude to that plugin
3. Claude can create canvases, set colors, paint strokes, and export to see results

Installation

Step 1: Install the Krita Plugin

Find your Krita resources folder:

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

Copy the plugin files:

krita_plugin/kritamcp.desktop  ->  [resources]/pykrita/kritamcp.desktop
krita_plugin/kritamcp/         ->  [resources]/pykrita/kritamcp/

On Windows, that's:

C:\Users\[YOU]\AppData\Roaming\krita\pykrita\kritamcp.desktop
C:\Users\[YOU]\AppData\Roaming\krita\pykrita\kritamcp\__init__.py

Enable the plugin in Krita:

1. Open Krita
2. Go to Settings > Configure Krita > Python Plugin Manager
3. Find "Krita MCP Bridge" and enable it
4. Restart Krita

Verify it's working:

• Open Krita
• The plugin starts an HTTP server on port 5678
• Test: Open browser to http://localhost:5678/health
• Should see: {"status": "ok", "plugin": "kritamcp"}

Step 2: Add the MCP Server

For Claude Code:

claude mcp add krita-mcp -- uv run --with fastmcp --with httpx fastmcp run /path/to/server.py

For Claude Desktop (add to config):

{
  "mcpServers": {
    "krita-mcp": {
      "command": "uv",
      "args": ["run", "--with", "fastmcp", "--with", "httpx", "fastmcp", "run", "/path/to/server.py"]
    }
  }
}

Tools

Example Session

Claude: krita_new_canvas(800, 600, "Evening Sky", "#1a1a2e")
        -> Created canvas: 800x600, background: #1a1a2e

Claude: krita_set_color("#b8a9c9")
        -> Color set to #b8a9c9

Claude: krita_set_brush("Soft", 30)
        -> Brush set: preset=Soft, size=30

Claude: krita_stroke([[100, 300], [200, 280], [300, 290], [400, 270]])
        -> Stroke painted with 4 points

Claude: krita_get_canvas("sky_v1.png")
        -> Canvas saved to: ~/krita-mcp-output/sky_v1.png

Claude: [views image, decides to adjust]

Claude: krita_undo()
        -> Undone

Claude: krita_set_brush("Soft", 50, 0.5)
Claude: krita_stroke([[100, 300], [200, 285], [300, 290], [400, 275]])

The Creative Loop

1. Paint - Use stroke, shape, fill commands
2. Export - krita_get_canvas() saves current state
3. View - Read the PNG file to see what was painted
4. Adjust - Undo, change settings, paint more
5. Repeat

This is different from code-based art. You're actually painting and seeing and responding to your own work.

Configuration

The plugin saves exports to ~/krita-mcp-output/ by default. Edit CANVAS_OUTPUT_DIR in krita_plugin/kritamcp/__init__.py to change this.

The HTTP server runs on port 5678 by default. Edit SERVER_PORT in the same file to change this.

Troubleshooting

"Cannot connect to Krita"

• Make sure Krita is running
• Check that the plugin is enabled (Settings > Configure Krita > Python Plugin Manager)
• Verify HTTP server: http://localhost:5678/health

Plugin doesn't appear in Krita

• Check file locations match exactly
• Make sure kritamcp.desktop is in pykrita folder (not inside kritamcp subfolder)
• Restart Krita after copying files

Commands timeout

• Krita might be busy with another operation
• Try again after a moment
• Check Krita's Python console for errors (Settings > Dockers > Log Viewer)

License

MIT