Agent Chrome

Let your AI coding agent use your real, signed-in Chrome. It works in its own tab group while you watch.

Agent Chrome lets your AI coding agent use the Chrome you already have open, signed in to your accounts. Ask your agent to check a dashboard, test your web app behind a login, fill in a form or reproduce a bug, and it does it in your browser. It works with any agent that supports the Model Context Protocol (MCP). How it works Agent Chrome is a small extension plus an MCP server. Your agent connects to the server (hosted at agentchrome.dev, or fully local on your computer), and the server passes each step to this extension. One command sets up the agents on your computer: curl -fsSL https://agentchrome.dev/install | sh You can also add https://agentchrome.dev/mcp as a custom connector in any AI chat app that supports MCP. You stay in control - Each agent works in its own named, colored tab group, apart from your tabs. - While it works, the page shows a blue border, a status bar with a Stop button and the agent's cursor. - Pause or stop any agent from the toolbar, block sites, or approve each new site. - Agents are told to ask you before buying, sending or deleting anything. What agents can do - Open, switch and close tabs, and read page text and structure. - Click, type, press keys, scroll, hover, drag and drop, fill forms and upload files. - Handle alert, confirm and leave-page dialogs. - Read console messages and network requests while testing a site. - Take screenshots, including long full-page images, and record MP4 or GIF clips. - Work across several Chrome profiles, such as Work and Personal. Privacy Page content passes through to your agent and isn't stored on our servers. Files an agent saves on purpose (screenshots, clips) are kept for 7 days. Keys are stored only as hashes. Agent Chrome is open source under the MIT license: https://github.com/ShadowWalker2014/agentchrome Made by Circo, Inc.