Connect your AI agent to your personal browser with auto-connect

Matthias Rohmer

José Luis Zapata

Connect your AI agent directly to your active browser instance with the auto-connect feature. This allows your agent to bypass sandbox limitations by inheriting your current tabs, browser extensions, and live application state.

Inheriting your workflow context lets your agent move from writing code to debugging your live browser state without losing session data. This is useful when you have already reached a specific application state (such as a complex checkout flow or a manually reproduced error) and want your agent to take over the technical investigation without re-executing those steps manually.

Consider the following before enabling auto-connect:

Security and privacy: When auto-connect is active, your agent has access to all data in your browser profile, including open tabs, session storage, local storage, cookies, and other data surfaced through JavaScript APIs.
Data privacy assurance: The Chrome DevTools for agents server is a local process and does not send your browser data, session tokens, or telemetry to Google.
Trust and prompts: Use this mode only with agents you trust and be careful with the information you include in prompts.

Prerequisites

Before you use the auto-connect feature, ensure you have the following:

Chrome 144+: Auto-connect requires modern debugging protocols found in current versions of Chrome.
Correct Chrome channel: Your agent connects to Chrome Stable by default. To use Canary or Beta, specify the channel in your configuration.
Remote debugging enabled: Manually enable the bridge between your browser and the agent.
Correct configuration: You must update the MCP configuration to include the --autoConnect flag.

Set up auto-connect

To configure the auto-connect feature in your MCP server, do the following:

In your running Chrome browser, navigate to chrome://inspect/#remote-debugging and enable remote debugging.

Update your MCP configuration to include the --autoConnect flag:

{
  "mcpServers": {
    "chrome-devtools": {
      "command": "npx",
      "args": ["chrome-devtools-mcp@latest", "--autoConnect"]
    }
  }
}

Prompt your agent to interact with your open Chrome app. When Chrome asks for permission to allow the session, click Allow.

Your agent now uses the MCP server to interact with your running Chrome application.

Use cases for auto-connect

Auto-connect lets you use tools or states that can't be used on sandboxed Chrome sessions. For example, instead of asking your agent to sign in, which often fails, you can navigate to the page yourself and sign in. Then, ask your agent to take over.

The following sections include some examples of use cases for the auto-connect feature.

Debug authenticated dashboards

Internal tools and private dashboards often sit behind complex single sign-on (SSO) or virtual private network (VPN) layers. A sandboxed browser usually requires a sign-in that your agent can't bypass. With auto-connect, your agent inherits your active session.

Example prompt:

I have my company's staging dashboard open in the next tab. Can you look at the
User Analytics chart, find the SVG element for the Friday data point, and tell
me why the tooltip isn't appearing on hover?

Example agent execution: Your agent identifies your existing Chrome process, finds the active tab, and inspects the accessibility tree to locate the elements. It may also use JavaScript APIs to evaluate specific conflicts.

Live-fix your current tab

Avoid manually reproducing errors. Instead of translating reproduction steps into a new prompt for a sandboxed agent, you can ask your agent to fix the bug on the active page where you already encountered it.