Chrome Relay CLI
Claude Code, Codex and Cursor agents work in your everyday Chrome with real logins, in background tabs, with an audit trail.
- 194 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The agent acts in your signed-in Chrome sessions: every site you are logged into is reachable.
- The service types passwords, codes and passkeys from 1Password; a wrong goal can send, delete or pay.
- The decider model receives up to 1500 characters of page text and the `--data` values.
Install
- Works in
Manual install
git clone https://github.com/aindeev/agent-chrome-relay
cd agent-chrome-relay
./build
./chrome-relay setupThe first README option: build from source on a Mac with Go 1.25+. The extension comes from the Chrome Web Store, and `setup` links the Claude Code skill.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The agent acts in your signed-in Chrome sessions: every site you are logged into is reachable.
- The service types passwords, codes and passkeys from 1Password; a wrong goal can send, delete or pay.
- The decider model receives up to 1500 characters of page text and the `--data` values.
This is third-party code. Review the repository files before installing.
What it does
Chrome Relay gives a coding agent your everyday Chrome together with signed-in sites. Agent tabs open inactive in a collapsed Agents group, and focus and windows do not move. Three parts work together: a Chrome extension, a background service on the Mac and the `chrome-relay` CLI. The `chrome-relay do` command takes a whole goal, for example inviting a teammate, then fills fields, clicks buttons and checks the `--until` condition itself. Step mode (`open`, `snapshot`, `click`, `fill`) answers in tens of milliseconds. Passwords and codes from 1Password are typed by the service, and the agent never sees them.
Who it is for. Mac users with Claude Code, Codex or Cursor whose agents need sites behind a login: dashboards, mail, CRM, cloud consoles.
Good fit when
- An agent needs to walk through an internal admin that only your account can open.
- A multi-step goal, such as an invitation form, repeats with different data.
- You are away from the Mac and approve sensitive actions from an iPhone with Face ID.
- An audit trail is needed, listing agent actions in the browser.
Not a fit when
- You work on Windows or Linux: the service and build target macOS.
- A clean browser without logins is enough: ordinary Playwright or Chrome DevTools MCP will do.
- The site must never be opened under your real account.
Example request
Open our dashboard through Chrome Relay, invite ada@example.com as an admin and wait for the text Invitation sent.Limitations
macOS with Google Chrome only; the build requires Go 1.25 or newer. The binary is built and signed on your Mac: no prebuilt one is issued. The extension is in the Chrome Web Store by direct link (unlisted), installed in each profile, and Chrome shows a debugging banner while it works. The password manager is 1Password only for now. The iPhone app is on TestFlight and awaiting App Store review. Quick decisions are made by an external model (TypeSafe Jev or Claude Haiku).
How to disable. Run `chrome-relay uninstall`, then remove the extension in each Chrome profile. The audit trail and saved flows live in `~/.chrome-relay/`.
FAQ
Does the Chrome window jump to the front?
No. Agent tabs open inactive, and commands that would raise the window are blocked.
Does the agent see my passwords?
No. `chrome-relay` types 1Password values, and the page snapshot shows only dots.
What does `chrome-relay do` do with dangerous buttons?
It hands control back to the agent with exit code 2 when a button such as delete, pay or send is needed that the goal does not name. `--allow` permits it.
Is 1Password required?
No. The core works without it: tabs, Google sign-in hops and the audit trail. 1Password, passkeys and the iPhone are optional.
Why does Chrome mention debugging?
This is expected: while agents work Chrome reports that the extension started debugging the browser. No remote-debugging port is opened.
Discussion
No questions or reviews yet.
Related
CLI toolsAn agent drives your real logged-in Chrome: a CLI, MCP and an extension with a separate tab group per session
A skill for Amp, Claude Code and Cursor: the agent drives your live Chrome with its open tabs and signed-in accounts through a single debugging toggle
A skill connects an agent to your Chrome through an extension and CLI for reading pages, managing tabs and verifying actions.
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
Surf drives your own Chrome from the command line: an agent opens pages, reads the accessibility tree, clicks and logs requests
A Claude Code plugin that drives your Chrome over the DevTools Protocol: tabs, forms, page extraction and screenshots with no third-party dependencies
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Chrome Relay (this card) | no | 194 | ||
| chrome-use | no | 257 | a CLI, MCP and extension with a tab group; Chrome Relay adds one-command goals through `do`, iPhone approvals and an audit trail | |
| Chrome DevTools MCP | yes | not required | 55,199 | the official MCP for debugging and traces of a live Chrome; Chrome Relay targets work with your logins in the background |
README in short
Each agent sees only tabs it opened. Step mode and `do` are written to the audit trail, and field values are not recorded.