agent-desktop
Reads and manipulates desktop applications via the operating system's accessibility tree.
- Type
- MCP
- Transport
- stdio
- Open source
- Yes
- GitHub Stars
- ★ 4.0k
- Source
- mcp-github
- Repository
- github.com/vercel-labs/agent-desktop
Overview
Agent Desktop is a tool built with Rust that inspects the true UI structure of any application through the operating system's accessibility tree and performs actions on it. This ensures stable references and safe retryable operations, rather than relying on pixel-based guessing. The tool supports macOS, Windows, and Linux and can be installed via npm. Ideal for AI agents requiring reliable interactions in desktop environments.
Capabilities
- ▪Read the UI structure of desktop applications
- ▪Manipulate UI elements by reference
- ▪Supports macOS, Windows, and Linux
- ▪Stable references without pixel matching
Use cases
Setup
npm install -g agent-desktop
This information was compiled by AI from public sources and may contain inaccuracies — please refer to the source.
FAQ
How to get permissions?
macOS requires Accessibility permissions: `agent-desktop permissions --guide`.
How to get started?
Run `agent-desktop snapshot --app Finder -i` to capture interactive elements, then manipulate them by reference.
Related skills
interview-coach-skill
Interview Coach skill covering the entire job search lifecycle from resume optimization to mock interviews.
skillpack
Package AI capabilities into local agents for use in Slack and Telegram.
harness-anything
AI agent control center supporting automation for WPS/MS Office, Zotero, Photoshop, and other software.
i-have-adhd
Optimize AI output to be more concise and action-oriented, tailored for users with ADHD.
work-iq
Query Microsoft 365 data, including emails, meetings, documents, and more, using natural language.
nostr-vpn
A private mesh VPN for humans and agents, supporting local keys and signed registration.