Files
openwork/packages/docs/computer-use.mdx
mintlify[bot] 765508cc15 Update computer use title and requirements
Mintlify-Source: dashboard-editor
2026-03-25 18:46:41 +00:00

43 lines
2.1 KiB
Plaintext

---
title: "OpenWork Use: Controlling your browser"
description: "How to setup the relevant skills for doing computer-use"
---
Computer-Use in OpenWork currently works as a browser automation in Google Chrome. It does not yet mean full desktop control of Ubuntu, macOS, or Windows apps.
With the right skills, your OpenWork agent will be able to runs browser actions and take screnshots as proof of having completed said actions.
If you connect to a remote OpenWork server, that remote machine also needs to have Chrome installed.
Today, the supported path is `Control Chrome`, which uses Chrome DevTools MCP.
## Requirements
- Google Chrome on the machine running the session
- In `Settings > Extensions > Control Chrome`
- In `Settings > Skills > enable `chrome-mcp-bootstrap`
## Run the first task
1. Open `Settings > Extensions`.
2. Add `Control Chrome`.
3. Start a new chat.
4. Click `Automate your browser` or run `/browser-setup`.
5. Ask OpenWork to open `https://example.com` and read the title.
6. Confirm you see Chrome tool steps and the response `Example Domain`.
Once you've verified this works, OpenWork can open pages, click, type, inspect page text, run small page scripts, and take browser screenshots.
Not supported today:
- arbitrary desktop apps outside Chrome (as restrictions around computer use heavily depends on your OS)
- OS-level window management or terminal or file-manager automation (again, depends on OS restrictions)
- a ready-made Docker image with Ubuntu desktop plus OpenWork plus Chrome as a full computer-use worker
## Fix common setup issues
- If the chat says Chrome MCP is not ready, reconnect `Control Chrome` in `Settings > Extensions`.
- If Chrome tool calls fail before a page opens, confirm Chrome is installed on the machine running the session.
- If you want OpenWork to attach to the Chrome you already use, enable remote debugging at `chrome://inspect/#remote-debugging` and then turn on `Use my existing Chrome profile`.
- If the chat fails before any browser tool runs, fix the model/provider first; browser automation only starts after the session can run a normal prompt.