Cua
The Short Version
Cua is open-source infrastructure for computer-use agents, AI agents that control full desktops (macOS, Linux, Windows). It provides sandboxes, SDKs, and benchmarks to build, train, and evaluate agents that click, type, scroll, and operate apps through a CLI and MCP surface. It’s built to work with agents like Claude Code, Codex, and OpenClaw.
This is advanced agent infrastructure, not a simple app. It’s for developers experimenting with desktop automation and agent research.
What Cua Is Best For
- Computer-use automation: let agents operate desktop apps and GUIs
- Agent infrastructure: sandboxes, SDKs, and benchmarks in one framework
- MCP workflows: run as an MCP server, daemon, or one-shot command
- Automation experiments: test background automation and agent control
Practical Use Cases
- Build desktop computer-use agents
- Automate GUI workflows
- Let agents interact with apps through CLI or MCP
- Test background automation
- Experiment with agentic desktop control and evaluation
Safety Note
Computer-use agents can take real actions on your machine. Avoid giving them access to sensitive apps, payments, credentials, or destructive operations without human approval. Run them in sandboxes or virtual machines and keep a human reviewing what they do.
Honest Limitations
- Advanced and technical: not a beginner app
- Requires setup and compute: sandboxes and VMs need resources
- Needs safety boundaries: computer-use agents must be supervised
Alternatives Worth Knowing
- OpenClaw, agent framework that works with computer-use setups
- Browser Use, agents that operate the browser specifically
- Claude Code, coding agent that can pair with computer-use infrastructure
Continue learning
Explore related guides, tools, workflows, and prompts that help you go deeper into this topic.
See how this tool fits into a workflow
Browse step-by-step AI workflows that use ChatGPT, Claude, Gemini, and other tools.
Frequently Asked Questions
What is Cua best for?
Cua is best for developers and AI builders experimenting with computer-use agents, AI agents that can control desktop apps. It provides open-source sandboxes, SDKs, and benchmarks to train and evaluate agents that operate macOS, Linux, and Windows through CLI and MCP.
Is Cua open source?
Yes. Cua is open-source under an MIT license, developed at github.com/trycua/cua. It's free to use, though running it requires technical setup and compute.
Are computer-use agents safe to run?
Treat them carefully. A computer-use agent can click, type, and operate apps on your machine, so you should avoid giving it access to sensitive apps, payments, credentials, or destructive actions without human approval. Use sandboxes or virtual machines and keep a human in the loop.
Last updated: