Run WebLoom 24/7. Any OS.
Drop one markdown into your AI, it configures the whole pipeline: headless Chrome, auto-start on reboot, cookies, cron schedules. Windows, macOS, Ubuntu, RHEL, Alpine — anywhere Chrome runs.
How it works
WebLoom now ships four cross-platform MCP tools that handle everything an Autopilot install needs, using OS-native mechanics:
launch_headless_chrome— spawns a headless browser on the target host. Auto-detects Chrome / Chromium / Edge / Brave.register_autostart— registers OS-native auto-start. Windows Task Scheduler / macOS launchd / Linux systemd--user.schedule_task— cron-style schedules. Windows schtasks / Unix crontab.autopilot_status— diagnostic. What's running, what's scheduled, health of each session.
You don't have to invoke them directly. Drop the setup guide markdown into your AI chat and it will use the tools to configure everything on whatever OS you have.
The primary path — one markdown into your AI
1. Install WebLoom MCP (if you haven't already)
One-click from Claude Code / Cursor / any MCP client. Grab it from the registry as io.github.webloomhq/engine — full guide at /docs/install.
2. Download the setup guide
# Or right-click → save as curl -O https://webloom.run/install/autopilot-setup.md
3. Drop it into your AI chat
Drag autopilot-setup.md into Claude Code / Cursor / Continue. Say what you want:
> Set up WebLoom Autopilot on this laptop, headless mode, no schedules yet. > Set up WebLoom Autopilot on my Vultr VPS at 1.2.3.4 (SSH key at ~/.ssh/vultr). > I want it posting to X and LinkedIn every Tuesday at 15:00 UTC. > Set up WebLoom Autopilot on my Windows Server. Just headless + auto-start, > I'll wire schedules myself later.
The AI reads the guide, asks any clarifying questions, then uses the four MCP tools to do everything. Cross-platform via platform.system(). Works on Windows, macOS, Ubuntu, Debian, RHEL, Fedora, Alpine, Arch.
The Ubuntu speed-run (one command, no chat)
If you know you want it on a fresh Ubuntu VPS and don't need Claude to hold your hand:
curl -sSL https://webloom.run/install/autopilot.sh | sudo bash
That script installs Chrome + Python + engine + systemd unit + a webloom-autopilot CLI shim in ~5 minutes. Same end result as asking the AI to do it, minus the flexibility for a different distro.
What the AI does (the flow it follows)
If you're curious what the guide tells the AI to do, here's the outline:
- Ask you where to install (this machine or SSH into a remote).
- Ask what you want it to do (which Thread, what cadence, or just headless + auto-start for now).
- Run
autopilot_statusfor a baseline. - Run
launch_headless_chrome, confirm the CDP endpoint is answering. - Run
register_autostartso it survives reboots. - Walk you through cookie handoff (
export_profileon your everyday session,import_profileon the headless one) for whichever sites you want to drive. - Install the Threads via
webloom_install. - Run
schedule_taskfor each cron entry you want. - Run
autopilot_statusagain to verify.
OS-specific notes the AI will ask you about
Windows
User-scope auto-start uses Task Scheduler ONLOGON. It restarts on next login after you close your session. For always-running server mode, ask for scope="system" — needs admin.
macOS
launchd LaunchAgents run in your login session. Same behavior as Windows-user.
Linux server without a login session
User-scope systemd needs sudo loginctl enable-linger <user> to run without a user shell. Otherwise use scope="system". The AI will ask.
Cookie handoff — how logged-in sessions get to the target
Fresh headless Chrome has no cookies. To drive X, LinkedIn, etc. it needs your existing authenticated session. WebLoom ships export_profile + import_profile for exactly this:
# On your everyday session (chat): > export_profile domain=x.com out=~/x-cookies.json # Move the file to the target (via scp / rsync / your usual means) scp ~/x-cookies.json user@1.2.3.4:/tmp/ # On the target session (chat, drives the headless Chrome): > import_profile path=/tmp/x-cookies.json session=headless
Re-export whenever the site expires the session. X ~30 days, LinkedIn ~14, most SaaS 90+.
Debugging: hosted trace viewer
Drop a trace_view before or after any fragile step. Returns a hosted URL you can open from anywhere:
> trace_view label="after LinkedIn post" # → https://webloom.run/trace/<id> — screenshot + DOM + metadata, 7-day retention
Cost + sizing
- Cheapest usable box: $5/mo, 1 GB RAM, 1 vCPU (any VPS provider). Chromium headless with a single tab fits comfortably.
- Comfortable box: $10-12/mo, 2 GB RAM. Handles 3-5 tabs simultaneously, vision_check overhead.
- WebLoom Pro: $19/mo. Break-even at 1 hour of your time saved per month.
Uninstall
Ask the AI to undo everything it set up — it knows because autopilot_status reports every entry it registered. Or manually:
- Windows:
Unregister-ScheduledTask -TaskName 'webloom-headless' -Confirm:$false·Remove-Item -Recurse ~/.webloom -Force - macOS:
launchctl unload ~/Library/LaunchAgents/webloom-headless.plist·rm -rf ~/.webloom ~/Library/LaunchAgents/webloom-headless.plist - Linux user:
systemctl --user disable --now webloom-headless.service·rm -rf ~/.webloom ~/.config/systemd/user/webloom-headless.service
Next steps
See the Cookbook for concrete recipe patterns you can drop into cron. Session-cookie best practices are in the Authoring guide. Support: dev@webloom.run.