Chrome Debug MCP — model-driven browser automation

Overview

The chrome-debug MCP server launches an isolated headless Chrome instance to enable AI-driven browser inspection and automation (take_snapshot, click, type_text, navigate, etc.) without requiring manual browser setup.

Getting started

Enabling browser automation

1. Navigate to Settings → MCP.

2. Find the chrome-debug row and tap Start.

3. In any chat in that project, the browser automation tools will become available to the model.

Pairing with the Inspector

By default, the Inspector tab and the chrome-debug MCP server operate independently:

• Inspector connects to a user-configured debug endpoint (default: http://127.0.0.1:9222).

• chrome-debug launches its own isolated Chrome instance for the model's actions.

  If you want both the Inspector UI and the model to share the exact same browser window, start Chrome with --remote-debugging-port=9222 and set the MCP server argument to:

--browser-url=http://127.0.0.1:9222

Behavior

• Headless by default — Chrome launches with an isolated temporary profile and no window, making it compatible with servers, containers, and desktop environments alike.

• Custom executable path — if Chrome is in a non-standard location, you can add --executablePath=/path/to/chrome in the server arguments.

• Tool safety — browser actions follow your project's tool authorization permissions (Ask, Allow, or Off).

Related

• MCP — managing Model Context Protocol servers.

• Inspector — mobile DevTools inspection tab.

• Tool authorization — tool execution permissions.