Desktop Automation Python Packages

cua-driver

Scale computer-use 2.0 with open-source drivers, cross-OS fleets, and benchmarks for training, evaluation, and data generation.

cua-bench

Scale computer-use 2.0 with open-source drivers, cross-OS fleets, and benchmarks for training, evaluation, and data generation.

cua-train

Scale computer-use 2.0 with open-source drivers, cross-OS fleets, and benchmarks for training, evaluation, and data generation.

cua-auto

Scale computer-use 2.0 with open-source drivers, cross-OS fleets, and benchmarks for training, evaluation, and data generation.

cua-sandbox

Scale computer-use 2.0 with open-source drivers, cross-OS fleets, and benchmarks for training, evaluation, and data generation.

a11y-computer-use

a11y-computer-use: accessibility-first computer use for AI agents. Pruned a11y-tree snapshots, element refs instead of pixel coordinates, a safety layer, an MCP server and CLI. macOS, Windows, Linux, and Chromium over CDP.

cua

Scale computer-use 2.0 with open-source drivers, cross-OS fleets, and benchmarks for training, evaluation, and data generation.

pymacos

A Python library for automating and controlling macOS.

crabagent

AI Knowledge Work Platform — Chat when you need answers, Work when you need results, Computer Use when the task is on your desktop. One agent across terminal, browser, and native macOS apps.

cua-computer-server

Scale computer-use 2.0 with open-source drivers, cross-OS fleets, and benchmarks for training, evaluation, and data generation.

cua-cli

Scale computer-use 2.0 with open-source drivers, cross-OS fleets, and benchmarks for training, evaluation, and data generation.

orbit-cua

Playwright but for any app on any OS

cua-som

Scale computer-use 2.0 with open-source drivers, cross-OS fleets, and benchmarks for training, evaluation, and data generation.

adobepy

Python SDK and Rust broker for Adobe desktop automation through UXP and CEP bridges, connecting AI agent tools and DCC-MCP adapters to Photoshop, Illustrator, InDesign, Premiere Pro, and After Effects.

openadapt

Compiles a demonstrated GUI task into a program that reports VERIFIED only if an independent check agrees. pip install openadapt; openadapt flow tutorial. Compiler lives in openadapt-flow.

dsh-cua

Windows computer-use MCP server for dsh (DeepSeek Harness) — client-neutral, works with any stdio MCP client. Accessibility-first element actions, skyshot text trees, guarded raw input, and a cross-session arbiter that yields to the human.

macos-computer-use-kit

AX-first computer use for AI agents on macOS with optional Jev (TypeSafe System One) semantic guards: calibrated target/input judgments before an irreversible action, decisions kept in code. Accessibility-tree targeting, window-scoped input, clipboard-safe paste, read-back verification. Ships a pip CLI, a pi package and a DeepSeek Harness plugin.

kwin-mcp

MCP server for Linux desktop GUI automation on KDE Plasma 6 Wayland -- 30 tools for virtual and live sessions

sassymcp

One MCP server, 270 tools, replaces 75+ — files, shell, GitHub, Android, desktop vision, security & forensics, persistent memory. Smart context-saving tool loading, safe-delete guardrails, and an orphan-proof self-healing process supervisor. Works with Claude Desktop, Cursor, Windsurf, and any MCP client.

extra-desktop

The Open-Source Astra 6 for your PC. Your AI can see, click, type, navigate, and get real work done on Windows.

branchkit

Python SDK for building BranchKit plugins

omni-computer-use-mcp

Windows computer-use MCP server - 1:1 replica of Claude Desktop's tool surface, for the Claude Code CLI or any MCP client. Screenshots, mouse, keyboard, multi-monitor. Full tier everywhere: type into IDEs, launch loose dev builds, let the agent test the app you're building.

wayland-mcp-kurojs

MCP server for Wayland desktop automation. Screenshot capture with VLM analysis, pointer and keyboard control, and action chaining for any MCP client.

oswright

MCP server for Windows desktop automation. Incremental perception instead of a screenshot per step: measured 16.9x fewer tokens per action than the alternatives, with the benchmarks to check it.