desktop-api

Cross-platform Python library for desktop automation: capture screens/windows and control mouse & keyboard. Works on macOS, Windows, and Linux.

Built on pyautogui, pygetwindow, and mss.

Installation
Quick Start
Examples
API Reference
Example Scripts
Safety Tips

Installation

# Install from PyPI
pip install desktop-api

# Or install from source
# Clone and install
git clone https://github.com/thedenk/desktop-api.git
cd desktop-api

python -m venv .venv
source .venv/bin/activate  # Linux/macOS
# .venv\Scripts\activate   # Windows

pip install -e .

Platform-Specific Requirements

Platform	Additional Setup
macOS	Grant Screen Recording and Input Monitoring permissions when prompted. Install PyObjC: `pip install pyobjc-core pyobjc-framework-Quartz pyobjc-framework-Cocoa`
Windows	No extra dependencies. Run as Administrator if automating elevated apps.
Linux	X11 only (Wayland not supported). Install: `sudo apt install scrot python3-xlib`

Quick Start

from desktop_api import DesktopController, WindowNotFoundError

# Create a controller
controller = DesktopController()

# Find and capture a window
try:
    window = controller.find_window("Safari")
    screenshot = controller.capture_window(window)
    screenshot.save("safari.png")
except WindowNotFoundError:
    print("Safari is not open!")

Examples

Creating a Controller

The DesktopController is your main entry point for all automation tasks.

from desktop_api import DesktopController

# Basic usage
controller = DesktopController()

# With safety options
controller = DesktopController(
    fail_safe=True,   # Move mouse to top-left corner to abort (default: True)
    pause=0.1         # Wait 0.1s between actions (default: 0.0)
)

Tip: Use pause=0.1 to give GUIs time to respond between actions.

Finding Windows

List All Open Windows

from desktop_api import DesktopController

controller = DesktopController()

# Get all visible windows
windows = controller.list_windows()

for win in windows:
    print(f"{win.title}")
    print(f"  Position: ({win.left}, {win.top})")
    print(f"  Size: {win.width}x{win.height}")
    print(f"  Active: {win.is_active}")
    print()

Find a Specific Window

from desktop_api import DesktopController, WindowNotFoundError

controller = DesktopController()

# Find by partial title match (case-insensitive)
try:
    window = controller.find_window("Chrome")
    print(f"Found: {window.title}")
except WindowNotFoundError:
    print("Window not found!")

# Find with exact title match
window = controller.find_window("Google Chrome", exact=True)

# Find without auto-activating the window
window = controller.find_window("Chrome", activate=False)

# Case-sensitive search
window = controller.find_window("Chrome", case_sensitive=True)

Activate and Refresh Windows

# Bring window to foreground
controller.activate_window(window)

# Refresh window geometry (after resize/move)
window = controller.refresh_window(window)
print(f"New position: ({window.left}, {window.top})")

Taking Screenshots

Capture Entire Screen

from desktop_api import DesktopController

controller = DesktopController()

# Capture primary monitor (full virtual screen)
screenshot = controller.capture_screen()
screenshot.save("fullscreen.png")

# Capture specific monitor (0 = all, 1 = first, 2 = second, etc.)
monitor1 = controller.capture_screen(monitor=1)
monitor1.save("monitor1.png")

Capture a Window

from desktop_api import DesktopController, WindowNotFoundError

controller = DesktopController()

try:
    window = controller.find_window("Notes")
    
    # Basic window capture
    screenshot = controller.capture_window(window)
    screenshot.save("notes.png")
    
    # Capture with padding around the window
    screenshot_padded = controller.capture_window(window, padding=20)
    screenshot_padded.save("notes_with_border.png")
    
    # Activate window before capturing (ensures it's visible)
    screenshot = controller.capture_window(window, activate=True)
    
except WindowNotFoundError:
    print("Notes app is not open!")

Capture a Region

from desktop_api import DesktopController

controller = DesktopController()

# Capture specific area: (left, top, width, height)
region = controller.capture_region((100, 100, 800, 600))
region.save("region.png")

# Using dict format
region = controller.capture_region({
    "left": 100,
    "top": 100,
    "width": 800,
    "height": 600
})

Mouse Operations

All mouse methods accept relative_to parameter for window-relative coordinates.

Click

from desktop_api import DesktopController

controller = DesktopController()
window = controller.find_window("Notes")

# Click at absolute screen position
controller.click(500, 300)

# Click relative to window (100px right, 80px down from top-left)
controller.click(100, 80, relative_to=window)

# Right click
controller.click(100, 80, button="right", relative_to=window)

# Triple click (select paragraph)
controller.click(100, 80, clicks=3, relative_to=window)

# Double click
controller.double_click(100, 80, relative_to=window)

Move Mouse

# Move instantly
controller.move_mouse(500, 300)

# Move with animation
controller.move_mouse(500, 300, duration=0.5)

# Move relative to window
controller.move_mouse(100, 100, relative_to=window)

Drag

# Drag from point A to point B
controller.drag(
    start_x=100, start_y=100,
    end_x=300, end_y=200,
    duration=0.3,
    relative_to=window
)

# Draw a line in a drawing app
controller.drag(50, 50, 200, 200, relative_to=window)

Low-Level Mouse Control

# Manual press and release (for custom drag operations)
controller.mouse_down(100, 100, relative_to=window)
controller.move_mouse(200, 200, relative_to=window, duration=0.5)
controller.mouse_up(200, 200, relative_to=window)

Scroll

# Scroll down (negative = down, positive = up)
controller.scroll(-3)

# Scroll up
controller.scroll(5)

# Scroll at specific position
controller.scroll(-3, x=100, y=100, relative_to=window)

Keyboard Operations

Type Text

from desktop_api import DesktopController

controller = DesktopController()

# Type text instantly
controller.type_text("Hello, World!")

# Type with delay between characters
controller.type_text("Slow typing...", interval=0.05)

# Type with newline
controller.type_text("Line 1\nLine 2\n")

Send Hotkeys

# Save file (Cmd+S on Mac, Ctrl+S on Windows/Linux)
controller.send_hotkey("command", "s")  # macOS
controller.send_hotkey("ctrl", "s")     # Windows/Linux

# Select all
controller.send_hotkey("command", "a")

# Copy
controller.send_hotkey("command", "c")

# Paste
controller.send_hotkey("command", "v")

# Undo
controller.send_hotkey("command", "z")

# Multiple modifiers
controller.send_hotkey("command", "shift", "s")  # Save As

API Reference

DesktopController Methods

Method	Description
`list_windows()`	List all visible windows
`find_window(query, exact=False, case_sensitive=False, activate=True)`	Find window by title
`activate_window(target)`	Bring window to foreground
`refresh_window(target)`	Get updated window geometry
`capture_screen(monitor=0)`	Screenshot of monitor
`capture_window(target, activate=False, padding=0)`	Screenshot of window
`capture_region(region)`	Screenshot of area
`click(x, y, button="left", clicks=1, relative_to=None)`	Mouse click
`double_click(x, y, relative_to=None)`	Double click
`move_mouse(x, y, duration=0.0, relative_to=None)`	Move cursor
`drag(start_x, start_y, end_x, end_y, duration=0.2, relative_to=None)`	Drag operation
`mouse_down(x, y, button="left", relative_to=None)`	Press mouse button
`mouse_up(x, y, button="left", relative_to=None)`	Release mouse button
`scroll(clicks, x=None, y=None, relative_to=None)`	Scroll wheel
`type_text(text, interval=0.0)`	Type string
`send_hotkey(*keys, interval=0.0)`	Send key combination

WindowHandle Properties

Property	Description
`title`	Window title
`left`, `top`	Position (top-left corner)
`width`, `height`	Dimensions
`right`, `bottom`	Computed edges
`is_active`	Whether window is focused
`handle`	Native window handle
`pid`	Process ID

Example Scripts

`examples/demo.py`

Basic window capture and click demo:

python examples/demo.py --window "Safari" --output screenshot.png

`examples/clicker.py`

Hotkey-based auto-clicker (requires pip install pynput):

# Hold Shift to click at 15 CPS
python examples/clicker.py --cps 15 --hotkey shift

# Toggle mode with Space key
python examples/clicker.py --cps 10 --toggle-hotkey space

`examples/dummy_agent_loop.py`

Template for AI agent automation loops:

python examples/dummy_agent_loop.py --window "Notes" --iterations 5

Complete Example: Automate Notes App

"""Full example: Open Notes, type text, save file."""
from desktop_api import DesktopController, WindowNotFoundError
import time

controller = DesktopController(fail_safe=True, pause=0.1)

try:
    # Find the Notes window
    notes = controller.find_window("Notes")
    print(f"Found: {notes.title} at ({notes.left}, {notes.top})")
    
    # Take a "before" screenshot
    before = controller.capture_window(notes)
    before.save("before.png")
    
    # Click to focus the text area (adjust coordinates for your app)
    controller.click(200, 200, relative_to=notes)
    time.sleep(0.2)
    
    # Type some text
    controller.type_text("Hello from desktop-api!\n")
    controller.type_text("This is automated input.\n")
    
    # Save with Cmd+S (or Ctrl+S on Windows/Linux)
    controller.send_hotkey("command", "s")
    time.sleep(0.5)
    
    # Take an "after" screenshot
    after = controller.capture_window(notes)
    after.save("after.png")
    
    print("Done! Check before.png and after.png")
    
except WindowNotFoundError:
    print("Please open the Notes app first!")

Safety Tips

Keep fail_safe=True – Moving mouse to top-left corner aborts automation
Use pause parameter – Gives GUIs time to respond between actions
Test with non-destructive actions first – Verify coordinates before automating
Add confirmation logic – For destructive operations, add explicit checks
Use activate=False – When you only need window metadata, not focus

Troubleshooting

Issue	Solution
`WindowNotFoundError`	Check if window is open and title matches
Permission denied (macOS)	Grant Screen Recording & Input Monitoring in System Preferences
Wayland not working (Linux)	Use X11 or XWayland session
Clicks going to wrong position	Use `refresh_window()` to update geometry after window moves
Typing not working	Ensure target window has keyboard focus

TheDenk/desktop-api

README