Your logins
Signed in like you are.
AI agents work inside your logged-in accounts (Gmail, GitHub, X, your bank), signed in exactly like you are.
Banking
Inbox
Docs
ISOLATED
Codex , flights
Claude , email
Browser for AI agents
The Missing Browser for Claude, Cowork & Codex
BrowserClaw is a local, open-source browser your favourite AI agents (Claude, Cowork, Codex, Cursor) drive using the accounts you're already signed into.You can actually trust what it does, because you watched it happen. Every session is recorded, replayable, and yours to change. It all runs on your machine.
GitHub Stars
10,000+
Data
On your machine
Chrome Import
One click
Compatibility
Any AI agent
Platforms
macOS, Windows
License
Free & Open Source
Your AI is smart. But it can't press the buttons.
Every AI assistant sounds capable in chat, then falls apart the moment it needs to actually open a tab, log in, and click through. BrowserClaw fixes that: be more productive with agents doing browser work in parallel, learn faster as it opens tabs for you, and trust the output by watching in real time with instant replay when it's done. Not a headless driver with no real logins. Not a cloud browser on someone else's machine. Yours, local, and watchable.
[ 01 ]What it does
A browser built for your AI to actually use.
We built the four things any agent needs to work in a browser. All shipped today.
Live cockpit
Watch every step.
A dashboard shows every running agent in real time.
Codex
Done12 toolsAudit trail
Every action logged.
Every dispatch and screenshot is a row on your disk, yours to grep or delete.
navigate→read→act→submit
Session replay
Rewind any session.
Watch what went wrong, then tell your agent what to do next time.
00:4203:16
Works with any AI agent.
- Claude Code
- Cursor
- Codex
- VS Code
- Zed
- GitHub Copilot
Powered by MCP, the open standard for AI integrations
[ 02 ]What BrowserClaw does
How your AI actually uses the browser
Step 01
Connect in one click.
Link BrowserClaw to Claude Code, Cowork, Cursor, or Codex via MCP. No config files, no CLI.
Step 02
Your agent browses live.
Ask in your terminal or IDE. The agent opens tabs, navigates, clicks, and reads the web signed in as you. Watch it happen in real time.
Step 03
Run several agents at once.
A cockpit dashboard on your new-tab page shows every agent working in parallel: which site, which tool, how many steps. Each runs in isolated, toggle-grouped tabs.
Step 04
Replay any session.
Every run saves as a full-screen, scrubbable video. Rewind, find the mistake, and coach the agent for next time.
Your stuff stays yours.
Your work stays uninterrupted.
Every agent opens its own tabs to work in and never touches the ones you already have open. It cannot close the doc you are writing, cannot steer the tab you are actively using, cannot step on your live window. You keep working while it works next to you.
Everything runs on your machine.
Session history, screenshots, replays, and settings are files on your disk. Nothing goes to a dashboard we own. Open source, top to bottom.
[ 03 ]Compare
How BrowserClaw compares.
| Feature | BrowserClaw | Others today |
|---|---|---|
| Agents use my real logged-in accounts | Yes | No, sandbox only |
| My AI can use a real browser | Yes | Some. Most use a spec. |
| I can watch my AI live | Yes | No |
| Run multiple agents in parallel | Yes | No |
| Named agent support (Claude, Cowork, Codex, Cursor) | Yes, one-click | Manual config |
| I can rewind a session and coach the agent | Yes | No |
| My AIs stay in their own tabs | Yes | No |
| Runs on my machine, no cloud | Yes | Some |
| Works with any AI agent | Yes, 6 one-click | Manual config |
| Free and open source | Yes | Some |
| Every action logged locally | Yes | No |
Get started in two minutes.
Download the browser.
Free for macOS and Windows.
macOS
WindowsConnect your AI.
Open BrowserClaw, choose your harness, click Connect.
Give your AI a task.
Just ask. Your agent reaches for BrowserClaw whenever the task hits the web.
$ claude
> Book me the cheapest
flight to London.
> opening BrowserClaw...[ 04 ]FAQ
Questions.
Everything you need to know about BrowserClaw.
BrowserClaw is a free, open-source browser your AI agents drive using your logged-in accounts. Your AI opens its own tabs inside it, and you watch, approve, and audit every step from a dashboard on your new-tab page.
Yes. BrowserClaw is completely free and open source under the AGPL-3.0 license. You bring your own AI provider keys.
Any AI agent that supports MCP, which is essentially every serious AI coding tool today. Claude Code, Cursor, Codex, VS Code, Zed, and GitHub Copilot install with one click. Anything else connects with a URL.
Yes. Open BrowserClaw, click Connect on the Claude Code row in the MCP install board, and Claude Code has a browser.
Yes. Codex is supported through the same one-click install as every other harness. No config file editing.
Yes. Claude Cowork connects through the same one-click MCP install as Claude Code, Cursor, and Codex. Any agent that speaks MCP can drive BrowserClaw.
BrowserClaw runs entirely on your own machine. It is not a cloud browser and not a headless driver. Your logins, sessions, screenshots, and replays are local files, and nothing goes to a server we own.
A headless browser has no real logins, so agents never touch your actual accounts. BrowserClaw is a full browser signed in as you, so agents do real work, and you can watch every step live and replay it afterward.
Not required. BrowserClaw is a standalone browser you install on macOS or Windows. If you also use BrowserOS as your daily human browser, the two products live side by side.
Everything runs on your machine. Session history, screenshots, replays, and settings are files on your disk. Your logins stay in your browser profile, same as any browser you use. Nothing goes to a dashboard we own. Open source, top to bottom.
Yes. Every agent gets its own set of tabs, tracked separately. Codex and Claude Code can each work on their own tasks at the same time without stepping on each other.
No. Every agent opens its own tabs to work in and never touches the ones you already have open. It cannot close the doc you are writing or steer the tab you are actively using. You keep working while it works.
Yes, that is the point. Agents drive BrowserClaw using the sessions you already have, so they automate your real work instead of poking a sandbox with no context. Per-agent profile isolation is on the roadmap if you want to keep them apart.
Yes. Every session is saved as a scrubbable video. Rewind, spot the moment things went sideways, and give the agent a better instruction next time.
macOS and Windows. System requirements match Google Chrome or Safari.
Yes. BrowserClaw is fully open source. The source code lives at github.com/browseros-ai/BrowserOS.
Give your AI your browser.
Free. Open source. Yours. Runs on your machine.
AGPL-3.0 license * 10k GitHub * Works with any AI agent