claude-workman
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@claude-workmanclick the 'Continue' button"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
claude-workman — computer use for Linux, as an MCP server
claude-workman is an open-source MCP server that gives an AI assistant human-mode control of a Linux desktop — it takes a screenshot, reads the accessibility tree, then moves the mouse, clicks, types, and switches windows. It works with Claude, Claude Code, and any other MCP client.
Named after the Walkman: a small, portable thing that just plays. Point it at a display and it works.
pip install claude-workman
claude mcp add --scope user workman -- python -m workman.serverWhy accessibility-tree clicking beats pixel clicking
Most desktop automation guesses at pixel coordinates and breaks the moment a layout shifts, a font renders differently, or a window moves. claude-workman fuses two signals on every turn — a screenshot and the AT-SPI2 accessibility tree — so an action can be addressed semantically:
"click the push button named Continue"
…and land on that button's center, wherever it moved to. This is the same principle behind
macOS AXUIElement-based computer use, ported to Linux via AT-SPI2 (gi.repository.Atspi).
Approach | Addressing | Survives layout change | Used here |
Pixel-only automation (raw xdotool, image matching) | guessed | ❌ brittle | fallback only |
Accessibility-tree automation | role + name → element | ✅ robust | default |
Related MCP server: computer-use-linux
Tools
Tool | What it does |
| Capture the display (or a sub-rectangle) as PNG; |
| Actionable elements as |
| Find an element by role + name and click its center — no pixel guessing |
| Turn on AT-SPI tree export (and silence the screen reader — see safety) |
| Windows with id, name, pid, geometry |
| Raise a window by id or name; minimizes the frontmost blocker first |
| Pointer actions at pixel coordinates |
| Keyboard input — |
| Visual click cursor overlay: a ring + click ripple showing exactly where the agent is acting |
Watch what it clicks
show_cursor starts a transparent, click-through overlay that draws a cyan ring at the action
point and an expanding ripple on every click, with a label. It exists for human oversight —
you can watch an agent work and interrupt it, instead of guessing what it just did.
Requirements
Linux with X11, plus:
sudo apt install xdotool ffmpeg gir1.2-atspi-2.0 python3-gi
# at-spi2-core is normally already running on GNOME/KDEffmpeg— screen capture (x11grab)xdotool— mouse and keyboard inputpython3-gi+gir1.2-atspi-2.0— the accessibility tree (import isgi.repository.Atspi, notpyatspi)Pillow(optional) — only formax_dimscreenshot downscalingGTK 3 (optional) — only for the
show_cursoroverlay
Install
pip install claude-workman # from PyPI
# or from source:
git clone https://github.com/AdvisorAGI/claude-workman.git
cd claude-workman && pip install -e .Usage
Run the stdio MCP server:
python -m workman.serverRegister it with an MCP client — for Claude Code:
claude mcp add --scope user workman -- python -m workman.serverTarget a specific X display with WORKMAN_DISPLAY (falls back to DISPLAY, then :0):
WORKMAN_DISPLAY=:1 python -m workman.serverThe accuracy loop
screenshot— andenable_accessibility+accessibility_treewhen precision mattersLocate — prefer an element by role + name; fall back to pixels
Act —
click_elementwhen possible, elseclick/type_text/press_keyscreenshotagain to verify before anything irreversible
Speech hazard (handled)
Exporting the accessibility tree from Chromium and Electron apps requires the desktop
screen-reader flag — and that same flag makes Orca read the screen aloud through
speech-dispatcher → espeak-ng. enable_accessibility therefore stops Orca immediately after
enabling, using an exact process-name match (pkill -x orca, never pkill -f orca, which
self-matches its own shell command and aborts).
GTK applications export their trees with toolkit-accessibility alone, so the web flag is
opt-in.
FAQ
What is claude-workman?
An open-source MCP server that lets an AI assistant control a Linux desktop the way a person does — screenshot, locate, click, type, verify — with accessibility-tree targeting so clicks hit real UI elements instead of guessed coordinates.
How is this different from browser automation?
Browser tools (Playwright, Puppeteer, CDP) drive a web page. claude-workman drives the whole desktop: native apps, terminals, file managers, settings dialogs, and browsers alike.
Does it work with Claude Code?
Yes. Register it with claude mcp add --scope user workman -- python -m workman.server. It
works with any MCP client, not only Claude.
Does it support Wayland or macOS?
Not yet — X11 today. A Wayland backend and a macOS AXUIElement backend are the natural next steps, and contributions are welcome.
Is it safe to let an agent control my desktop?
Treat it like handing over mouse and keyboard. Use show_cursor so you can see every action,
prefer a dedicated display or VM for unattended runs, and keep the verify-by-screenshot step
before anything irreversible.
Why is it called Workman?
After the Walkman — a small portable thing that just plays. Point it at a display and it works.
Does it solve CAPTCHAs?
No. It is a general desktop-control server, not an anti-bot bypass tool.
Contributing
Issues and pull requests are welcome — see CONTRIBUTING.md. Good first areas: a Wayland backend, a macOS AXUIElement backend, richer AT-SPI role coverage, and tests.
License
MIT © 2026 Atmosphere AI — see LICENSE.
Keywords: MCP server, Model Context Protocol, computer use, desktop automation, Linux automation, X11 automation, AT-SPI2 accessibility, GUI agent, AI agent desktop control, xdotool alternative, screenshot automation, Claude MCP connector, agentic computer use.
This server cannot be installed
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Related MCP Servers
- Alicense-qualityDmaintenanceA desktop automation MCP server that enables AI agents to interact with Linux environments through screenshots, window inspection, and input simulation. It provides tools for mouse control, keyboard input, and screen capture using xdotool and XDG Desktop Portals.MIT
- Alicense-qualityAmaintenanceControls a real Linux desktop from any MCP host. Reads accessibility trees, takes screenshots, and drives clicks, scrolls, and keystrokes across GNOME, KDE/KWin, Hyprland, i3, and COSMIC.671370MIT

gui-userofficial
Alicense-qualityCmaintenanceAn MCP server for external computer-use that launches, observes, and interacts with any X11 application via AT-SPI2 accessibility tree and xdotool input injection.MIT- Alicense-qualityBmaintenanceEnables MCP clients to control macOS via accessibility and screen recording, providing tools to list apps, observe UI, click, type, press keys, and scroll.MIT
Related MCP Connectors
Eyes and hands on real Windows PCs — observe, click, type via Glasswarp API.
Hosted real Google Chrome MCP with per-user persistent state. Navigate, click, type, screenshot.
Access Kernel's cloud-based browsers and app actions via MCP (remote HTTP + OAuth).
Latest Blog Posts
- Who's Calling? MCP Hosts Are an Identity Blind Spot (And the Spec Knows It)By Om-Shree-0709 on .mcpAgent IdentityOAuth 2.1
- Your AI Chatbot Just Exposed Your CEO's Salary to an InternBy Om-Shree-0709 on .Agent IdentityMCP SecurityOAuth Delegation
- Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)By Om-Shree-0709 on .Agentic AiPrompt InjectionWebAssembly
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/AdvisorAGI/claude-workman'
If you have feedback or need assistance with the MCP directory API, please join our Discord server