An integrated terminal in your browser that lets you use AI coding agents to control live tabs (works with Claude Code, Codex, Cursor, Hermes, and more).
agent-terminal-demo.mp4
Connect your agents to Chrome with interactive browser automation, in Chrome’s side panel, on your live tabs.
Note: this project is experimental. Agents can control your live Chrome and run shell commands on your machine. Use with care.
This extension lets coding agents drive the page you’re looking at: open, click, fill, snapshot in your real Chrome, while you watch and intervene.
| Agent Terminal | IDE / Terminal.app alone | |
|---|---|---|
| Interactive browser + agent | Side panel interaction next to live browser tabs | Easy to spawn invisible tabs you never see |
| You stay in the loop | Same window (watch the agent, take over anytime) | Automation often runs off-screen |
| One-click agents | Claude / Codex / Cursor / Hermes / OpenClaw | Manual shells and PATH |
Prerequisites: Go 1.24+, Node.js 20+, Chrome.
git clone https://github.com/rishiskhare/agent-terminal.git
cd agent-terminal
go build -o agent-terminal .
./agent-terminal install
cd extension && npm install && npm run buildThen load the extension in Chrome:
- Open
chrome://extensions - Turn on Developer mode
- Click Load unpacked and choose the
extension/dist/chrome-mv3folder in this repo - Open the side panel from the Agent Terminal extension icon
This extension uses Vercel’s agent-browser CLI, so the extension never touches page DOM. Browser automation runs using normal CLI tools inside the PTY. Agent Terminal does not add heavy MCP servers or a custom automation stack.
It is a real terminal in Chrome’s side panel: your shell, local files, git, language toolchains, existing CLI/agent tools, and whatever else is on your PATH. By default, coding agents are guided to use agent-browser on your live Chrome tabs through an AGENTS.md file (custom instructions can be configured by directly editing this file).
Because it is just a terminal, you can use any browser automation you prefer: agent-browser, Playwright, Puppeteer, Selenium, Chrome DevTools Protocol, or another CLI/MCP you already trust.
- Browse freely: while your agent browses the web, you can browse other tabs. No need to keep the agent’s working tab in focus.
- Tabs: up to 8 terminals per Chrome window
- One-Click Agent Launcher:
+runs your last agent; dropdown menu for Claude, Codex, Cursor, Hermes, OpenClaw, or Terminal - Settings: shell, theme, font size, env, and Run doctor in Chrome extension options
- Session survival: close the side panel and reopen later; the agent is still running
- Isolated sessions: concurrent agents each get their own browser session
- Live Chrome only: attaches to your Chrome (never Chrome-for-Testing)
- Setup doctor: detects agent CLIs and sets up launchers plus
AGENTS.md
Agents attach to the Chrome you’re already using. They will not start Chrome-for-Testing. You can keep browsing other tabs; the agent does not need its tab focused to keep working.
- Turn on remote debugging once at
chrome://inspect/#remote-debugging(leave Chrome open) - Click Allow if Chrome asks after a restart
PRs welcome. For large changes, open an issue first. Keep the extension free of page host permissions, and keep browser automation on the user’s live Chrome.
This project is licensed under the MIT License. See LICENSE. Based on Tweety by Achille Lacoin.