Browser control for coding agents, a command line for your tabs, and a userscript loop from your editor straight into the browser.
LBcat is a browser extension plus a small local CLI called bctl. It does three jobs in the browser you already use,
with your logins:
- Browser control for agents. Claude Code, Codex, or any tool that can run shell commands uses
bctlto open tabs, read pages, click, type, and take screenshots, inside one agent window you hand it. Your other tabs stay out of reach. - A CLI for your tabs.
bctl human tabslists, focuses, and closes tabs across every window, as plain text or JSON, for scripts, launchers, and tab switchers. - Userscripts from disk. Write scripts in TypeScript or plain JavaScript, build them to
.user.js, andbctl script watchinstalls each build in the browser as soon as it is saved. No copy and paste, no install dialog. Updates land the same way.
LBcat is a fork of ScriptCat, so it is also a full Tampermonkey-compatible userscript manager. All three jobs need a native bridge between a local program and the extension, which upstream ScriptCat does not include, so we added it and ship our own signed build.
Status: free and open source. Tested on macOS. Linux should work: the tools build and install from source there, but we have not tested them on Linux yet, and desktop notifications are macOS-only for now. There is no Windows support yet.
- Your real browser. No separate profile and no logging in again. The agent sees the sites the way you do.
- A scoped agent window. You designate one window for agents. Commands never fall back to your personal window, and a personal tab is only reachable when you grant that exact tab.
- Task sessions and tab leases. Each task gets its own session and owns the tabs it opens, so several agents can share one browser without stepping on each other. You can pause a task and take a tab back at any time.
- A real dev loop for userscripts.
bctl script pushmatches installed scripts by@nameand@namespace, so each push updates the installed copy in place.push-allinstalls a whole folder, in one browser or all of them. - Firefox and Zen, not only Chromium. Chrome, Brave, Edge, Vivaldi, Firefox, Zen, and other forks are detected.
- Built for agents. Output is compact text by default and JSON on request, to keep token use low.
- Local only. The CLI talks to the extension over a Unix socket and Native Messaging. Nothing is sent to a server.
This is command targeting and task ownership, not a sandbox. Cookies and site state are shared with your profile, so only point agents you trust at it.
You need two parts: the extension and the local tools.
-
Firefox and Zen: install LBcat from Firefox Add-ons.
-
Chrome and other Chromium browsers: a Chrome Web Store listing is on the way. Until then, build it and load
dist/ext/chromeas an unpacked extension:pnpm install pnpm run build
On macOS:
brew install latentbyte/tap/lbcat
browser-control installThe first command installs browser-control (also linked as bctl), browser-control-daemon, and lbcat-shim. The
second registers the Native Messaging host for every browser it finds. Restart the browser afterwards.
Chromium browsers trust the LBcat Chrome Web Store ID. For an unpacked build, pass its ID from chrome://extensions:
browser-control install --extension-id <id>.
On Linux, or to build from source on macOS, you need Go 1.25 or newer and just.
This puts the tools in ~/.local/bin (add it to your PATH) and registers the Native Messaging host:
git clone https://github.com/latentbyte/lbcat
cd lbcat
just browser-control-install
brew install terminal-notifier # macOS only: click-to-approve notificationsbctl browsers # connected browsers, e.g. zen-01
bctl --browser zen agent-window create # the window agents may use
bctl --browser zen session begin --label "demo" # prints a handle such as zen:ab12cd34
bctl --session zen:ab12cd34 tabs create https://example.com --active=false
bctl --session zen:ab12cd34 page content <tab-id>
bctl --session zen:ab12cd34 page screenshot <tab-id> --output /tmp/page.png
bctl --session zen:ab12cd34 session end --close-tabsThe full command reference is in backend/browser-control/README.md.
Copy backend/browser-control/skills/browser-control into your
agent's skills directory, for example ~/.claude/skills/browser-control. The skill teaches the agent the session
workflow, the stopping conditions, and how to ask you for access.
Go programs can use the SDK in backend/browser-control/pkg/browsercontrol.
Everything ScriptCat does still works: Tampermonkey-compatible scripts, background and scheduled scripts, cloud sync, and the built-in editor. See the ScriptCat documentation for writing scripts.
To develop scripts on disk instead of in the built-in editor, point your build at a folder and watch it:
bctl script watch ~/my-scripts # install each .user.js as soon as it changes
bctl script push dist/my-script.user.js # one-off create or update
bctl script push-all ~/my-scripts --all-browsersSee docs/DEVELOP.md and docs/ARCHITECTURE.md. Common commands:
just check # tsc and eslint
just test # extension tests
just browser-control-test # Go testsThe extension is licensed under GPLv3, the same as ScriptCat, which it is built on. Thanks to the ScriptCat authors and contributors for the userscript engine.
The Go CLI, daemon, and SDK in backend/browser-control/ are original work licensed under
Apache-2.0.
