qz CLI
A Chrome extension with a CLI, MCP and a skill: Claude Code and Codex hand qz a web task that runs in your own browser
- 220 ★ on the repository
- Last commit
- Apache-2.0
Broad access
Why this level
- A task runs in your Chrome with active logins and cookies and can perform on sites any action available to you
- Page content goes to the model you choose, and a terminal agent can start tasks without your involvement
Install
- Works in
Manual install
npx skills add alibaba/qz-browser-agent -gInstall the qz extension in Chrome first. Then run npx qz-browser-agent@latest connect to pair, Node.js 22 or newer is required.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- A task runs in your Chrome with active logins and cookies and can perform on sites any action available to you
- Page content goes to the model you choose, and a terminal agent can start tasks without your involvement
This is third-party code. Review the repository files before installing.
What it does
qz works as a Chrome extension with a side panel. A task is stated in one sentence and qz runs it on sites where you are already signed in: it searches and reconciles data from several sources and builds tables, files and reports with source links. You connect your own model: DeepSeek, OpenAI, Claude, and for a ChatGPT account there is a sign-in through OpenAI Codex without an API key. For other agents the project ships the `qz-browser-agent` npm package with a command line and an MCP server, plus the `skills/qz-browser-agent/SKILL.md` skill. An agent such as Claude Code or Codex installs the skill, runs `connect` once to pair with the browser and then hands the whole web task to qz instead of individual clicks. Progress is visible in the side panel, and you can step in or stop the task at any moment. qz runs several tasks in parallel, and each can spawn subtasks. A task that worked can be saved as a skill with the phrase "Save this as a Skill" and repeated with one sentence.
Who it is for. People who ask Claude Code or Codex to collect data from sites behind a login or compare a dozen sources and do not want to set up a separate headless browser for it.
Good fit when
- You need a comparison table with sources: models, plans or specs from different sites
- The data sits behind a login and your regular Chrome already holds the session
- A weekly report, such as a list of fast-growing repositories, should be done once and saved as a skill
- A terminal agent needs a browser, while you want to see every step and be able to stop it
- The research is large and worth splitting into parallel subtasks
Not a fit when
- The agent runs in a cloud shell: the CLI must run on the machine with Chrome and the extension, and the cloud cannot reach a desktop browser
- Node.js is older than 22: pairing and the CLI require it
- You need a headless browser on a server with no user: qz is built around a visible Chrome and a side panel
- Pages cannot be passed to a model because of data requirements: the content goes to the model you choose
Example request
Use qz to compare the open-weight LLMs released in the last 6 months in a table with release date, parameters, context length, license and sourcesLimitations
- The CLI and pairing need Node.js 22 or newer. The agent and the browser must run on the same machine.
- The extension installs from the Chrome Web Store or manually from the release archive in developer mode. Automatic Chrome launch during pairing depends on the chosen profile.
- An API key for the chosen provider is required, while OpenAI Codex signs in through a ChatGPT account.
- The extension works with your sessions and cookies, so on sites where you are signed in a task can perform any action available to you.
How to disable. Disable or remove the qz extension on the `chrome://extensions` page. Remove the skill by deleting its folder from the agent's skills directory. If the CLI was installed globally, run `npm uninstall --global qz-browser-agent`.
FAQ
How do I connect qz to Claude Code or Codex?
Install the extension in Chrome, then run `npx skills add alibaba/qz-browser-agent -g` and `npx qz-browser-agent@latest connect`. After pairing tell the agent "use qz for this task". For MCP clients `npx qz-browser-agent@latest mcp` prints the server's HTTP address.
Where does my data go when qz works?
According to the project, task data is stored locally and page content goes only to the model you choose. You set the provider and key yourself in settings.
Can I install the extension without the Chrome Web Store?
Yes. Download the archive from the latest release, extract it, open `chrome://extensions`, turn on developer mode and pick the `qz-browser-agent-extension` folder with Load unpacked.
How is qz different from Playwright MCP?
Playwright MCP gives an agent individual browser actions, while in qz a whole task is stated in one sentence and runs in your Chrome with your logins. The comparison table in the README was written by the author.
Which models can I connect?
Your own providers, for example DeepSeek, OpenAI and Claude, with an API key. For a ChatGPT account there is a sign-in through OpenAI Codex without a key. The choice is made in the extension settings.
Discussion
No questions or reviews yet.
Related
CLI toolsSix skills for OpenCLI: an agent reads, searches and posts on dozens of sites through your Chrome and repairs adapters.
An agent drives your real logged-in Chrome: a CLI, MCP and an extension with a separate tab group per session
A skill connects an agent to your Chrome through an extension and CLI for reading pages, managing tabs and verifying actions.
An MCP server and Chrome extension give any agent 22 tools for your own logged-in browser, verifying the effect of every action.
A CLI and MCP server that lets an agent drive your real Chrome with its logged-in sessions: 103 commands across 36 platforms plus browser automation
A skill and CLI from Browser Use: the agent drives your real Chrome over the DevTools Protocol, logged in as you, and writes any missing helper itself
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| qz (this card) | no | 220 | ||
| Playwright MCP | yes | not required | 39,570 | gives an agent individual browser actions, while qz takes a whole task and runs it in the user's visible Chrome |
| agent-browser | yes | not required | 50,530 | a fast Rust CLI with accessibility snapshots, while qz works as an extension with a side panel and its own model |
README in short
The README shows the extension through example requests with screenshots: a trip plan, model comparison, saving a skill, parallel research and delegating from an agent. It then gives a four-step quick start, manual installation without the Chrome Web Store, a section on connecting outside agents through the skill and `connect`, and a comparison table with other browser agents.