SideButton
Open-source MCP server with knowledge packs, 40+ browser tools, and YAML workflow engine for AI agents.Read full description
sidebuttonLABSCO SUMMARY
The same browser layer, four ways in.
Most browser automation projects pick one integration and make you live with it. This one ships the same extension-plus-CLI core behind MCP, a skill, and SDKs in three languages, which means the choice of agent runtime stops being a lock-in decision. It drives your real Chrome through a native host rather than a headless copy, so sessions and logins are the ones you already have.
What it is
A browser automation layer that does not tie itself to one agent runtime, positioned as an open alternative to the Chrome browser-use capability shipped in Codex.app. Under the hood it pairs a Chrome extension with the open-browser-use CLI, and you reach it through MCP, an installable skill, or the JavaScript, Python and Go SDKs.
What you get
- Browser tools over MCP for tab listing, opening and claiming, navigation, CDP, action plans and cleanup
- The same capability as an installable skill for Codex or Claude Code, added with
npx skills addand updatable in place - Three SDKs over one CLI —
open-browser-use-sdkon npm and PyPI, and a Go package imported asobu - A one-command Chrome setup that registers the native host and opens the Web Store page for the matching extension, with a keyed beta ZIP as a fallback for manual installation
- Downloadable
.skilland.zippackages published in GitHub Releases
Requirements
Chrome, plus the CLI: npm i -g open-browser-use (Homebrew also works on macOS and Linux), then open-browser-use setup to register the native host and install the extension. For MCP, npx add-mcp "obu mcp" --name open_browser_use --all -g -y writes the config into every supported agent, or you point a client at the obu command with the mcp argument yourself. Licensed MIT.
Setup effort
Follow the published steps — npm i -g open-browser-use
USE CASES
01
Giving an agent your actual browser session
Register the native host and let the agent claim a tab in the Chrome you are already signed into, rather than driving a fresh headless profile.
02
Adding browsing to Codex or Claude Code in one command
Install the skill globally with npx skills add, then ask the agent to read a page and summarise it.
03
Keeping the same automation across agent runtimes
Configure the MCP server once with add-mcp across every supported agent config, so switching runtime does not mean re-plumbing browser access.
04
Scripting it outside an agent
Call the same CLI from the JavaScript, Python or Go SDK when the work belongs in a program rather than a conversation.
BEFORE YOU INSTALL
What to know first
Requires
Google Chrome
INSTALLATION
Connect your client
Server setup
TERMINAL
npm i -g open-browser-use
open-browser-use setupIn the Chrome Web Store page the setup command opens, install or enable the matching Chrome extension, then restart Chrome if needed.
MCP Clients
Any MCP client — add-mcp
ANY MCP CLIENT — ADD-MCP · TERMINAL
npx add-mcp "obu mcp" --name open_browser_use --all -g -yCodex
CODEX · TOML
[mcp_servers.open_browser_use]
command = "obu"
args = ["mcp"]Codex — skill
CODEX — SKILL · TERMINAL
npx skills add iFurySt/open-browser-use -g -a codex --skill open-browser-use -yClaude Code — skill
CLAUDE CODE — SKILL · TERMINAL
npx skills add iFurySt/open-browser-use -g -a claude-code --skill open-browser-use -y
