Automate logged-in Chrome with Codex or Claude Code

chrome-use is an open-source CLI and MCP server that connects to your existing Chrome through an extension and native messaging, retaining that profile’s signed-in sessions.

Install and verify

curl -fsSL https://raw.githubusercontent.com/leeguooooo/chrome-use/main/install.sh | sh
chrome-use extension install
chrome-use doctor

Finish extension setup using the installation guide. A doctor result does not establish which website account is signed in.

A reproducible read-only task

chrome-use --session seo-example open https://github.com/leeguooooo/chrome-use
chrome-use --session seo-example snapshot -i
chrome-use --session seo-example get title

Expected: the repository opens, snapshot returns page elements, and the title contains leeguooooo/chrome-use. No messages are sent or account data changed.

How do agents connect?

Shell-capable Codex and Claude Code use the CLI and load chrome-use skills get core. MCP-only clients use chrome-use mcp.

How does it compare with Playwright or browser-use?

Existing sessions, human takeover and multiple agents sharing one browser are chrome-use’s focus. Playwright supports programmed tests and existing-browser connections. browser-use supplies an agent framework. Compare actual connection paths rather than assuming other tools always start empty browsers.

Does it guarantee no CAPTCHA?

No. Fingerprint measurements do not guarantee site acceptance. The user handles login, CAPTCHA and second-factor verification.

MCP · Sessions · Comparison · Detector evidence

Try a real browser task

Give this task to Codex, Claude Code, or another agent with shell access. It searches public GitHub issues in your existing Chrome. You can watch the tab while your agent works.

Set up the CLI and Chrome extension · View the source

Use chrome-use to search https://github.com/leeguooooo/chrome-use/issues
for is:issue daemon. Load chrome-use skills get core first.
Use a dedicated session named first-task. Read snapshot -i for current
refs, fill the Search issues box, and submit the search with --observe.
Verify the query in the resulting URL and read the issue titles.
Report three matching titles and links. Do not create or edit any issue.
If Chrome needs login, permission, or human verification, let me handle it.

Recorded on October 7, 2026 in our existing signed-in Chrome: fill the issue query, submit, then read the filtered results. The 5.3-second clip shows browser actions only; it excludes installation, model reasoning, and the final written report. This public search also works when signed out, so the clip demonstrates browser control, not access to private data.

Check your first result

The resulting URL should contain q=is%3Aissue+daemon (equivalent URL encoding is fine). Your agent should return matching titles and links from the page. A command returning success by itself is not proof the task finished.

For a task that needs login, use a dashboard you own and start with reading only. Confirm the visible account before continuing. You can take over login or verification in the same Chrome window.

Stuck? Tell us which setup step failed, with your OS, CLI version and a redacted error. Leave out cookies, tokens and private page contents. Browser tasks run locally; this docs site uses analytics to measure visits and setup-link clicks.