Octop Browser CLI
Agent browser on direct CDP without Playwright: login profiles, a CLI, an MCP server and recording workflows into a skill
- 75 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The agent acts in the browser as you wherever the profile is signed in
- The `eval_js` action runs arbitrary JavaScript on the open page
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/octop-browserDetects 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 acts in the browser as you wherever the profile is signed in
- The `eval_js` action runs arbitrary JavaScript on the open page
This is third-party code. Review the repository files before installing.
What it does
Octop Browser launches a real Chromium with `--remote-debugging-port` and talks to it over the Chrome DevTools protocol itself, with no Playwright layer. The agent receives page elements as `ref` links from the live DOM tree, and those links stay valid after a re-render. The DOM tree is multi-level and compact, so a request spends fewer tokens. There are about twenty actions: navigate, click, type, fill a field, press a key, wait, scroll, screenshot, tab handling and JavaScript evaluation. Each is available as an `octop-browser` command and as an MCP tool. A named profile keeps logins on disk, so you do not sign in again. Workflow recording captures your actions in the browser and turns them into a draft skill with `record generate-skill`. Replay runs through the agent by the meaning of the steps, not by coordinates. The `skills/` folder holds ready skill files in English and Chinese.
Who it is for. Agent developers and testers who need a controllable browser with persistent logins and low token spend.
Good fit when
- The agent misses elements in Playwright after a re-render and you want refs from the live DOM
- You want to sign in to a service once by hand and then let the agent work in the same profile session
- A repeating action in a web interface should be recorded and turned into a draft skill
- You need a browser MCP server for Claude Code or Cursor started with one command
Not a fit when
- You need a ready end-to-end test suite with reports: this is a browser control tool with no test harness
- You need Firefox or WebKit support: only Chromium is launched
- You must keep the agent away from a browser holding your logins: the profile stores sign-in on disk
Example request
Open the sign-in page in the work profile, wait until I sign in, then open the reports section and take a screenshotLimitations
Python 3.11 or newer and a Chromium binary are required; `octop-browser install-browser` downloads it once and may bootstrap Playwright a single time for that, but it is not needed at runtime. A profile keeps sign-ins on disk and the agent acts in your name in services where you are signed in. `BROWSER_USE_IDLE_TIMEOUT_MINUTES` defaults to zero, so Chrome does not stop by itself. The author publishes no accuracy comparisons with other tools, so the lower token spend remains a README claim.
How to disable. Close the session with `octop-browser close-session --kill --profile <name>` and remove the package with `pip uninstall octop-browser`. Remove the MCP entry from the agent config separately. Profile data stays on disk; delete the profile folder if the logins are no longer needed.
FAQ
How is Octop Browser different from Playwright?
It does not use Playwright at runtime and talks to Chromium over CDP directly. Playwright is needed only once, to download the browser binary with `install-browser`.
How do I start the MCP server?
With `python -m octop_browser.mcp_server`, or `make mcp` from a clone. The transport is stdio and each action is its own tool: `browser_navigate`, `browser_click` and so on.
Where are logins stored?
In a profile on disk. You name it with `--profile`, for example `work`. Closing the session does not erase the sign-in, and the next command attaches to the same profile.
What does workflow recording do?
`record start` captures actions in the browser, `record generate-steps` shows the semantic steps and `record generate-skill` emits a draft skill `draft.skill.md`. `replay run` has the agent follow the meaning of the steps.
Are ready-made skills included?
`skills/octop-browser` holds an English SKILL.md and `skills/octop-browser-zh` a Chinese one. They describe safe use: refs in place of coordinates and named profiles for login state.
Discussion
No questions or reviews yet.
Related
CLI toolsA browser for agents and people with an MCP server: the agent works in tabs with your logins while you watch and can take over
A local Playwright driver for your everyday Chromium browser: the agent works in your live, logged-in profile through an extension and a relay
A Claude Code plugin with an MCP server and seven skills: the agent drives a real browser by writing Python code
A separate browser for agents with a skill and MCP: Claude Code and Codex work in your logged-in accounts, and every session is recorded for replay
An MCP and CLI that let an agent control your real Chrome, with its own logins and extensions, through Playwright
A fast Rust CLI for agent browser automation: accessibility snapshots with element refs, an MCP server and skills
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Octop Browser (this card) | no | 75 | ||
| Playwright MCP | yes | not required | 39,570 | works through Playwright |
| Browser Use | no | 120,809 | a CLI, skill and Python library on CDP with no recording into skills | |
| chrome-devtools-cli | no | 256 | a Rust binary for your already open Chrome with no profiles or recording |
README in short
The README presents a lightweight browser-use style tool with direct CDP. Tables of actions, CLI commands, workflow record and replay, the MCP server and a Python API reference follow. It states separately that Playwright is needed only for the initial Chromium download.