Chrome Control MCP

Lets AI agents drive your real Chrome over MCP, in tabs of their own with a visible cursor. Your mouse and keyboard stay yours.

Chrome Control MCP lets AI agents use your own Chrome -- with your sign-ins, your extensions and your profiles -- through the Model Context Protocol (MCP). Any MCP client works: Claude Code, VS Code, Cursor, Codex and others. This extension is half of it. The other half is the Chrome Control MCP server, a small program you install on your computer from GitHub (free, open source, MIT licence): https://github.com/RandyNorthrup/chrome-control-mcp What agents get: - Tabs of their own. Each agent opens its own tabs, grouped and framed in its own colour, and cannot reach into another agent's tabs or yours unless you hand one over. - A visible cursor. You see where an agent clicks and types; your own mouse and keyboard stay yours. - Fast, cheap observation: find an element by its text, read a page's text, take a screenshot or an accessibility snapshot with references to click. - Real input: clicks, typing, keys, drag and drop, file uploads without a file dialog, downloads, printing to PDF, recordings of a tab. - A clipboard per agent, kept apart from yours. - Several Chrome profiles, chosen per call ("work", "personal"). Privacy: what the extension reads from a page goes only to the Chrome Control MCP server on your own computer, through Chrome's native messaging. There is no account, no analytics and no server of ours; nothing is sent to the developer. Policy: https://github.com/RandyNorthrup/chrome-control-mcp/blob/main/docs/PRIVACY.md Click the toolbar button to stop every agent at once; click it again to let them back. Requires Chrome Control MCP server 1.13.0 or later.