Open Browser Use CLI
A CLI, Chrome extension, SDKs and MCP server that let an agent drive your real Chrome, whatever the agent runtime
- 421 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The agent drives your real Chrome profile with all its open sign-ins, forms and purchases
- The CLI and SDKs have no per-site policies, method allowlist or approvals, so all caution rests on the agent and the skill
- Any local process with access to the socket in `/tmp/open-browser-use/` can control the browser
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/open-browser-useDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The agent drives your real Chrome profile with all its open sign-ins, forms and purchases
- The CLI and SDKs have no per-site policies, method allowlist or approvals, so all caution rests on the agent and the skill
- Any local process with access to the socket in `/tmp/open-browser-use/` can control the browser
This is third-party code. Review the repository files before installing.
What it does
Open Browser Use connects an agent to a real Chrome profile. It consists of an MV3 Chrome extension, a local native messaging host written in Go, the `open-browser-use` command line (short name `obu`), SDKs for JavaScript, Python and Go and an optional stdio MCP server. The author calls it an open alternative to the Chrome Browser Use capability that appeared in Codex.app and keeps it independent of any single agent runtime. The commands cover the usual tab loop: `ping`, `info`, `tabs`, `user-tabs`, `open-tab`, `navigate`, `history` and `cdp` for direct DevTools protocol calls. A tab you already opened is taken under control with `claim-tab`, and when the work ends tabs are released or kept with `finalize-tabs`. Short multi-step scenarios are written as a line-oriented plan and run with `run`. A skill, `open-browser-use`, comes with the CLI and collects the rules: a unique session id per task, a tab group labeled OBU, reusing an already open tab instead of a duplicate, and asking the user before submitting forms, purchasing, deleting or uploading files. The same set is available as MCP tools, including `run_action_plan` and `finalize_tabs`. The repository also holds a package for an embedded Electron browser, not yet published to npm.
Who it is for. For developers and agent users who need automation of a real Chrome with their own tabs and sessions, without tying to one agent vendor.
Good fit when
- An agent must check a site or your local development in an already open Chrome tab where you are signed in
- You want a digest from several pages, for example to browse fresh Hacker News posts and pick the worthwhile ones
- You want one browser control layer for several agents: through the CLI, the JavaScript, Python or Go SDK, or MCP
- You want to record a multi-step tab scenario as a line-oriented plan and run it repeatedly
- You need low-level DevTools protocol commands to debug a page from a script
Not a fit when
- You need an isolated environment for untrusted code: the extension works in your real Chrome profile and is no sandbox
- You need per-site policies, a method allowlist and action approvals inside the tool: neither the CLI nor the SDKs have them, that is the upper layer's job
- You need a headless browser on a server with no Chrome and extension: the tool needs a real Chrome and the store extension
- You need to accept remote commands from outsiders: an integration must add its own authentication, audit log and confirmations
Example request
Use open-browser-use to check today's Hacker News and summarize the five most worth-reading postsLimitations
The tool gives an agent low-level control of a real browser and is not a security boundary. Per-site policies, allowed methods and approvals for risky actions must come from the layer that calls it. Sockets sit in `/tmp/open-browser-use/`, the `active.json` file is created with mode 0600, and access to your account on the computer equals access to the browser control channel. The author plans client tokens and client-process validation later.
You need Chrome with the extension from the Chrome Web Store, and before that the `open-browser-use setup` command, which registers the native messaging host. If the store page is unavailable, `setup beta` prepares a ZIP for manual install. The repository skill works outside Codex but does not use Codex.app helpers or interfaces. MCP commands for the embedded Electron browser go through a separate `obu iab mcp` route, and that package is not on npm yet.
How to disable. Remove the extension on the `chrome://extensions/` page, then run `npm uninstall -g open-browser-use`. The skill is removed with `npx skills remove open-browser-use -g`, and the MCP entry is deleted from the agent configuration.
FAQ
How do I connect Open Browser Use to Claude Code?
Install the CLI with `npm i -g open-browser-use`, run `open-browser-use setup`, install the Chrome extension, then add the skill: `npx skills add iFurySt/open-browser-use -g -a claude-code --skill open-browser-use -y`. For MCP, `npx add-mcp "obu mcp" --name open_browser_use --all -g -y` is enough.
How does it differ from Playwright?
Playwright launches a separate managed browser. Open Browser Use works in your everyday Chrome profile through an extension, sees your tabs and sessions and can take an already open tab under control.
Which SDKs exist?
The `open-browser-use-sdk` package on npm and PyPI for JavaScript and Python, and a Go module `github.com/ifuryst/open-browser-use/packages/open-browser-use-go`. In Python the import is `open_browser_use`, and in Go the package is named `obu`.
What if the extension page in the store is unavailable?
Run `open-browser-use setup beta`: it prepares a keyed ZIP of the extension for manual install from `chrome://extensions/`. An archive you already downloaded goes in with the `--zip` flag and an absolute path.
How do agent tasks avoid getting in each other's way?
Each task gets its own session id and a tab group labeled OBU. The fallback session `obu-cli` is meant only for manual runs, because it can reuse other tasks' tab groups.
Is it safe to give an agent access to Chrome?
It is access to a real profile with your sign-ins. The skill forbids reading cookies and passwords and requires asking you before forms, purchases and sending. The tool itself imposes no such rules, so the upper layer sets the order.
Discussion
No questions or reviews yet.
Related
CLI toolsAn agent drives your real logged-in Chrome: a CLI, MCP and an extension with a separate tab group per session
A Claude Code plugin with an MCP server and seven skills: the agent drives a real browser by writing Python code
CLI, skill and Python library for browser control: the agent clicks, fills forms and reads pages over CDP
OpenTabs
opentabs
An agent calls the internal APIs of web apps through your signed-in browser session, with no keys or OAuth. Chrome extension and CLI. Actions go in your name
A plugin and MCP server letting an agent drive your open, already logged-in Chrome through an extension with no debugging port.
An MCP server and Chrome extension give any agent 22 tools for your own logged-in browser, verifying the effect of every action.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Open Browser Use (this card) | no | 421 | ||
| Playwright CLI | yes | 15,133 | launches a separate managed browser instead of working in your profile | |
| Browser MCP | yes | not required | 7,345 | an MCP server and Chrome extension for driving an already open browser, without SDKs or a CLI |
| Open Computer Use | no | not required | 2,940 | by the same author: controls the whole desktop |
README in short
The description, in English and Chinese, presents the project as a neutral Chrome automation layer and links to the author's deep dive. A quick start follows, with CLI install through npm and Homebrew, Chrome setup and three ways to connect: SDKs, the skill and MCP. The embedded Electron browser and the MIT license are mentioned at the end.