Chrome Gemini CLI

CLI for AI-powered browser automation. Tell the agent what you want, and it autonomously navigates, clicks, fills forms, and extracts data.

Built with Gemini AI and Chrome DevTools MCP.

Demo

1. Go to Target and search for 'bookshelf'. 2. Apply these filters: Color = White, Material = Wood. 3. Set max price to $80. 4. Show only items that are available for order pickup. 5. List the links for all results.

Prerequisites

• Node.js >= 20.19.0
• Google Chrome browser
• Gemini API key

Installation

pnpm install
cp .env.example .env
# Add your GEMINI_API_KEY to .env

# Optionally, create your own config to override defaults
cp .chrome-geminirc.default.json .chrome-geminirc.json

Usage

Default Interactive Mode:

Starts a chat session.

npm start

Run a Task:

Executes a one-off task with default settings.

npm start -- "Find Honolulu's weather on Google"

Using a Preset:

Executes a task using a named preset for configured, non-interactive, or schema-driven tasks.

# This assumes a "product-scrape" preset exists in your config
# that sets a schema and disables interactive mode.
npm start -- --preset product-scrape "Get the top 3 laptops from Best Buy"

Scripting (Quiet Mode):

For clean JSON output suitable for piping or redirecting, use npm start --silent.

# Using a preset and piping output to a file
npm start --silent -- --preset product-scrape "Get the top 3 laptops" > output/laptops.json

Important: When redirecting output (>) or piping (|), always use npm start --silent to prevent progress logs from appearing in your output.

Configuration

Chrome Gemini CLI uses an opinionated configuration system where the config file controls all non-secret settings.

Configuration is resolved in this order (highest to lowest priority):

1. CLI Flags - For temporary, one-time overrides (presets only).
2. Project Config - .chrome-geminirc.json in the current directory (optional, overrides defaults).
3. Default Config - .chrome-geminirc.default.json (built-in, always present).

Environment Variables (.env file)

Environment variables are used strictly for secrets. All other configuration belongs in a config file.

# .env
GEMINI_API_KEY=your_key_here

Configuration Files

This is the standard way to configure the CLI.

Project Config (.chrome-geminirc.json):

Create this in your project's root directory. It controls how the tool should behave.

{
  "interactive": true,
  "maxSteps": 30,
  "model": "gemini-3-flash-preview",
  "requestDelayMs": 1000,
  "schema": "",
  "temperature": 1,
  "thinkingLevel": "medium",

  "presets": {
    "quick": {
      "maxSteps": 20,
      "interactive": false
    },
    "research": {
      "maxSteps": 50,
      "thinkingLevel": "high"
    }
  }
}

Configuration Fields:

• interactive - Enable interactive chat mode after task completion
• maxSteps - Maximum number of agent steps before stopping
• model - Gemini model to use (e.g., gemini-3-flash-preview, gemini-3-pro-preview)
• requestDelayMs - Delay in milliseconds between Gemini API calls to prevent rate limiting
• schema - Path to a TypeScript schema file for structured JSON output
• systemPrompt - Path to a markdown file with custom system prompt instructions
• temperature - Model temperature for response randomness (0 = deterministic, 2 = maximum creativity)
• thinkingLevel - Reasoning depth for model thinking (minimal, low, medium, high)
• presets - Named configurations that can be invoked with --preset <name>

The tool comes with .chrome-geminirc.default.json that provides sensible defaults. You can override any values by creating .chrome-geminirc.json.

CLI Options

The CLI is intentionally minimal, prioritizing configuration files.

• --preset <name> - Use a named preset.
• --help - Show usage information.

Using Presets

Presets allow you to create named configurations for common workflows. They are defined in the presets object for running automated tasks.

# Use the "quick" preset
npm start -- --preset quick "Find the price of an iPhone"

# Use the "research" preset
npm start -- --preset research "Discover the top 5 AI companies, and summarize the Wikipedia page of each company."

Note: Use npm start --silent to suppress all logs and get clean JSON output (perfect for piping/scripting).

Schemas

Schema files for structured output must:

• Be TypeScript (.ts) files
• Import Type from @google/genai
• Export a schema object with Gemini Type enums

Example (schemas/examples/price-schema.ts):

import { Type } from '@google/genai'

export const priceSchema = {
  type: Type.OBJECT,
  properties: {
    product: { type: Type.STRING },
    price: { type: Type.STRING },
    available: { type: Type.BOOLEAN }
  },
  required: ['product', 'price']
}

Add your own schemas in schemas/ directory (git-ignored). See schemas/examples/ for examples.