- Home
- Skills
- Workflow Automation
- Agent Desktop for Linux Pro
Works with the AI Agents you already use
Agent Desktop for Linux Pro
Everything in Agent Desktop for Linux, plus OCR clicks, set-of-marks vision, smart waits, macros, session reports, parallel desktops and an MCP server.
$12
Agent Desktop for Linux Pro
Example session with this skill installed
Accept the cookie banner on this site, find the "Download report" button and download it. Show me what you clicked and give me a report of the run.
- Read your context and instructions
- Compiled the agent desktop
- Generated the document
Done. I worked on my own private desktop, so your pointer and windows were never touched.
What I did
| Step | Command | Result |
|---|---|---|
| Start + audit trail | adc start --nested · adc journal start | Private display, every step logged with a screenshot |
| Open the site | adc browser https://example.com --name work | Agent-owned Chrome profile |
| Wait for load | adc wait-stable --ms 700 | Page settled in 1.4 s |
| Cookie banner | adc click-text "Accept all" | Found by OCR at (1032, 588), clicked |
| See the page | adc marks --app chrome | 23 numbered elements; #14 = button "Download report" |
| Download | adc click-mark 14 | Clicked |
| Check | adc downloads | ~/Downloads/report-2026.pdf (412 KB) |
| Report | adc report --out ~/adc-report.html | 7 steps, 0 errors |
Safety
adc guard on was active: a click on "Send" or "Pay" would have stopped and asked you first.
Next steps
- Open
~/adc-report.htmlto review every step with screenshots. - Save this flow with
adc macro start download-reportand replay it any time withadc macro play download-report.
report-2026.pdf
PDF · document
Example file from a real run - the skill writes it into your workspace.
Connects securely to your tools. The creator never sees your data.
About this skill
Your AI agent gets its own isolated Linux desktop, a private display with its own pointer, keyboard and clipboard, so it never touches your mouse or steals your focus. Pro adds the tools you need for real, repeatable automation: it can see and click anything on screen, wait correctly, replay workflows and show you exactly what it did.
Everything in the free edition
- Isolated desktop: a private Xephyr display on its own workspace, with its own focus, clipboard and D-Bus session.
- Accessibility engine: snapshot any app, then click, type, select and toggle by element reference (GTK, Qt, Chrome, Electron).
- Human-like input: a visible agent cursor glides to each element; typing uses the real keyboard, so web apps receive genuine events.
- Persistent browser profiles: sign in once, stay signed in.
- User hand-off: the agent asks you to complete logins, captchas and 2FA, then continues.
- Screen recording to MP4.
Pro features
- Set-of-marks vision:
adc marksdraws numbered boxes on every clickable element of a screenshot; the agent acts withclick-mark 7. A picture and exact targets in one step. - OCR:
find-text,click-textandwait-textfind any visible text, even inside buttons and inputs, in canvases, remote desktops or apps without accessibility. - Smart waits:
wait-stableandwait-changereplace fixed sleeps; they ignore a blinking caret. - Guard: with
adc guard on, actions such as send, pay, delete or publish (English and Spanish) are refused until the user confirms. - Macros with variables: record a workflow once and replay it with different data (
--var email=...). Elements are matched by role and name, so replays survive layout changes. - Journal and HTML report: every step logged with a screenshot, exported as a self-contained HTML timeline you can share.
- GIF export of any recording.
- Parallel desktops:
ADC_SESSION=nameruns several isolated desktops at once, for several agents or tasks. - MCP server:
adc mcpexposes the desktop as tools to Claude Desktop, Cursor or any MCP client; screenshots come back as images. - Clipboard and downloads of the agent desktop.
How it works
- Run
adc doctorto check dependencies. adc start --nestedstarts the agent's private desktop;adc journal startbegins the audit trail.- The agent observes with
snapshot,marksorsee, and acts with references, marks or text. - It waits with
wait-stable/wait-textand verifies the result. adc stopcloses everything gracefully;adc reportgives you the timeline.
Use cases
- Repetitive back-office work in web apps and desktop software without an API.
- Filling the same forms with different data through macros.
- Downloading reports or exporting data from dashboards behind a login.
- QA runs on desktop and web apps, with an HTML report and a GIF as evidence.
- Several agents working in parallel, each on its own desktop.
Known limitations
Linux with an X11 session only (no Wayland, macOS or Windows). OCR accuracy depends on text size and contrast. Apps without an accessibility tree rely on OCR and coordinates.
Requires an X11 session; no Wayland, macOS or Windows. OCR accuracy depends on text size and contrast. Apps without an accessibility tree rely on OCR and coordinates. System packages (Xephyr, openbox, tesseract) must be installed; adc doctor lists them.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 8 days ago
Requires
Permissions
File Scopes
Runs a private X display (Xephyr) on its own workspace. Writes screenshots and state to $XDG_RUNTIME_DIR/adc; agent browser profiles, macros and journals to ~/.local/share/adc (mode 0700); optional recordings to ~/Videos. Reads ~/Downloads only to list files the agent downloaded in the session. Network is only used by the browser the agent opens. Credentials are never typed by the agent: sign-ins are handed to the user.