| name | chrome-browser |
| description | Control the user's real Chrome browser with existing logins and cookies. Use for navigation, clicking, typing, screenshots, file upload, reading page content, and page-context JavaScript. |
| allowed-tools | Bash, Read, Write |
Chrome Agent Bridge
Use the chrome-browser CLI to control the user's actual Chrome session through the installed Chrome Agent Bridge extension.
Setup checklist
Before first use:
- Install the Chrome extension from the repo or Chrome Web Store.
- Build the Rust binaries.
- Install the native messaging host manifest.
- Verify with
chrome-browser ping.
Core commands
chrome-browser ping
chrome-browser navigate '{"url":"https://example.com"}'
chrome-browser getContent '{"format":"annotated"}'
chrome-browser click '{"text":"Sign in"}'
chrome-browser type '{"selector":"input[name=q]","text":"hello","submit":true}'
chrome-browser fillForm '{"fields":[{"selector":"#email","value":"a@b.com"}]}'
chrome-browser screenshot '{}'
chrome-browser evaluate '{"script":"return document.title"}'
chrome-browser scroll '{"y":500}'
chrome-browser uploadFile '{"selector":"input[type=file]","path":"/tmp/file.pdf"}'
Recommended workflow
chrome-browser pingchrome-browser navigate ...chrome-browser getContent '{"format":"annotated"}'- Use selectors or visible text from the annotated output for
click / type / fillForm
- Use
screenshot when visual confirmation matters
Notes
- This controls the real Chrome browser, not a headless copy.
- Authenticated sessions and cookies are preserved because actions happen inside the user's actual browser.
- Chrome Web Store pages themselves may restrict extension automation.
evaluate runs in the page's main world and can return JSON-serializable values.
Supported actions
pingnavigategetContentgetInteractablesclicktypefillFormwaitForscrollevaluatescreenshotlistTabsnewSessionsetActiveTabgetActiveTablistFramesuploadFiledropFilereloadforkkillForklistForksparalleltryUntil
