Browser automation · DWG 04
agbrowse
A browser automation CLI for AI agents
agbrowse lets AI agents drive Chrome from the command line. It observes the page, acts through stable references, and collects screenshot, console and network evidence. There is no long-running MCP server, so tool descriptions never tax every request.
Spec
- Install
npm install -g agbrowseagbrowse start- Category
- Browser automation
- License
- MIT
- Repository
- github.com/lidge-jun/agbrowse
- GitHub · npm
- 230+ stars120+ weekly downloadsSeptember 25, 2026 as of
Features
- FIG. 01
Reference-based actions
Click and type through references from a snapshot, so actions survive layout shifts.
- FIG. 02
Evidence on demand
Capture screenshots, console logs, network requests and DOM state with one command each.
- FIG. 03
Short-lived processes
Each command is a short Node process that reconnects to the same Chrome DevTools endpoint. No resident server.
- FIG. 04
Web AI sessions
Send prompts to logged-in ChatGPT, Gemini and Grok web UIs and bring the results back.
Use it when
- You want your coding agent to check the page it built in a real browser
- You would rather not spend context on MCP tool descriptions
- You need an agent to handle web work that requires a login
FAQ
Do I need an MCP server?
No. Every command runs on its own and prints only what you ask for.
Which browser does it use?
Chrome-family browsers controlled over the Chrome DevTools Protocol.
Which Node version?
Node.js 18 or later.