Browser Agent Bridge

Bridges your real Chrome profile to a local MCP server for AI agent control.

Browser Agent Bridge lets an AI agent that you run on your own computer, such as Claude Code or Claude Desktop, or a script you write, work in your real Chrome profile: the one with your accounts and cookies. How it works: this extension connects to a small program on your computer (at 127.0.0.1) that your agent starts. The agent asks it to read a page, click, type, scroll, take a screenshot, read the console or network activity, or run a bit of JavaScript, and the extension does it in a tab. Nothing is sent to the project, and the extension has no analytics. You need the other half too: add the bridge to your agent. For Claude Code run claude mcp add browser-agent-bridge -- npx -y browser-agent-bridge. For Claude Desktop, install the .mcpb file from the project's GitHub Releases. The extension then pairs with it by itself; there is no token to copy. You stay in control: A toolbar button and the shortcut Alt+Shift+K turn the bridge off at once. While it is off, every page call is refused. A separate switch takes JavaScript execution away from the agent. A block list keeps the agent off the sites you name. File uploads only work from folders you list, and the list starts empty. The bridge never opens local file pages or its own settings page. These settings are enforced inside the extension. The agent cannot change them. Please read before you install: an agent connected to the bridge can do in your signed-in tabs what you can do. Connect only agents you trust. Page content the agent reads goes to the AI service that agent uses, under that service's terms. Open source (AGPL-3.0). Source, security model and privacy policy: github.com/Capta1n-n9m0/browser-agent-bridge