Tools
browxai
Browser and app control for AI agents, over MCP.
Who it is for
Developers who give an AI agent a browser, a phone or a desktop app to drive, from Claude Code, Codex, Pi or a TypeScript script.
The agent gets compact tools to navigate, find, click, fill, read and take screenshots, with small structured results.
Proof
What it does today
Browsers, apps and desktop
Chromium, Firefox, WebKit, real Safari and Chrome on an Android device, plus native apps on an Android emulator and the iOS Simulator, and a running Electron app.
Your own Chrome
browxai launches your installed Chrome on one default profile per workspace. When a site shows a bot-protection challenge, browxai chrome login lets a person clear it by hand.
No evasion
browxai does not defeat or spoof bot detection.
Off until you turn it on
In-page code evaluation, files, the clipboard, password managers and plugins stay off until you enable them.
How to start
Get going
browxai is pre-1.0 and published on npm.
01
Run it
Run npx browxai on Node 22 or later.
02
Connect a client
Point an MCP client such as Claude Code or Codex at it, or call it from a TypeScript script.
03
Enable only what you need
Turn on only the extra tools you use.
Install
npx browxai
Prompt for your agent
Add browxai to this project as an MCP server. It runs with npx browxai on Node 22 or later. Connect it to this client and turn on only the extra tools we need.
The details
- Licence
- MIT
- Status
- Open source, released
browxai gives an AI agent compact tools to navigate, click, fill, read and take screenshots, from an MCP client such as Claude Code or Codex, or a TypeScript script. It drives five browsers, including real Safari and Chrome on an Android device, plus native apps on the Android emulator and iOS Simulator, and Electron apps. Risky tools stay off until you turn them on. browxai does not defeat or spoof bot detection.