Earth Agent Chrome Extension

Earth Agent is a Cursor-like AI agent for Google Earth Engine. It can be used right in your browser as a Chrome extension or through MCP server support. It helps you do anything related to Google Earth Engine automatically through chatting (write code, run analysis, debug errors, explain maps, and manage your environment). Hatched from sundai.club.

[!TIP] Quick Start: Install from Chrome Web Store, set up your API key (OpenAI/Anthropic/Google) or MCP server with Claude Code/Cursor/Zed/OpenCode, and say "Help me map NDVI for California".

Table of Contents

Features
Installation
MCP Server Support
Configuration
How to Use the Extension
Usage Examples & Prompts
Available Tools & Functions
Usage Tips
Troubleshooting & Help
Privacy and Security
Roadmap
Changelog
Contributing

Features

Chat interface for Earth Engine assistance
Knows Earth Engine Data Catalog as well as community dataset
Knows Earth Engine APIs
Help you write code, and run the code
Help debug the code
Help you explain the map
Planning and reasoning
Environment management tools (reset map/console, clear code)
Comprehensive agent testing framework with automated testing capabilities
Custom Instructions: Tailor the AI's behavior and responses to your specific needs. Work like AGENTS.MD.
Agent Profiles: Create and switch between different agent configurations for different tasks
MCP Server Support: Use Earth Agent tools directly from Claude Code, Cursor, or Zed via the Model Context Protocol.

Installation

Option 1: Install from Chrome Web Store (Easiest & Recommended)

Best for: Standard users who want to use OpenAI, Anthropic, or Google Gemini.

Visit the Earth Agent page on the Chrome Web Store.
Click "Add to Chrome".
The extension will be added to your browser and will appear in your Chrome toolbar.

[!NOTE] The Web Store version does not support local connections (MCP Server or Ollama) due to store policies. For those features, use Option 2.

Option 2: Download from GitHub Releases

Go to the Releases page
Download the latest earth-agent-extension.zip
Extract the zip file to a folder on your computer
Open Chrome and navigate to chrome://extensions/
Enable "Developer mode" (toggle in the top right)
Click "Load unpacked" and select the extracted folder
The extension will appear in your Chrome toolbar

Option 3: Install from Source

Clone the repository
Install dependencies with npm install
Build the extension with npm run build
Load the unpacked extension from the dist directory in Chrome

Configuration

[!TIP] You do not have to do this if you only want to use Earth Agent as MCP server. If you want to use it directly in your browser without external agents tools, you need to do this.

API Keys

Click the Earth Agent extension icon in Chrome
Go to Settings
Configure your AI providers keys. we support OpenAI, Anthropic, Google, Z.AI, and Custom OpenAI-Compatible Providers. You need input at least one of them to use Earth Agent.
Go back to the chat page, select your preferred model
Start chatting with Earth Engine!

Custom Instructions (AGENTS.md)

Just as software developers use files like CLAUDE.md, GEMINI.md or AGENTS.md to define project-specific rules for their AI coding assistants, Earth Agent allows you to set Custom Instructions to align the AI with your scientific or policy goals. Think of it as giving the agent a "Standard Operating Procedure" for your research. This is powerful for:

Scientific Rigor: e.g., "Always use the 'Cloud Score+' dataset for cloud masking in Sentinel-2 images."
Policy & Communication: e.g., "Draft all explanations as policy briefs suitable for decision-makers, avoiding jargon."
Methodology Standardization: e.g., "Always use median compositing instead of mean for annual composites."
Coding Conventions: e.g., "Use functional programming styles in GEE (e.g., map()) over client-side loops."

To configure:

Open Extension Settings
Use Custom Instructions to set global preferences

Agent Profiles (Custom Agents)

This allows you to create your own agent with specific system prompts and tool access for specific workflows or use cases. Each agent profile can have its own:

Custom System Prompt: Define exactly how the agent should behave
Tool Access: Enable or disable specific tools for focused tasks

Example Profiles:

"Strict Coder": Focused purely on writing optimized code, with no chit-chat.
"Tutor": Explains every step in detail for learning.
"Analyst": Specialized in statistical analysis and visualization.

To configure:

Open Extension Settings
Scroll to Profiles to create new agent personas

How to Use the Extension

You can use this extension in two ways: through the MCP Server or inside the Earth Agent side panel. MCP servers expose the same tools as the Earth Agent assistant. Check MCP Server Support for details on using it as an MCP server. The following section will show you how to use it inside the Earth Agent side panel.

Getting Started

After configuration:

Open Google Earth Engine Code Editor in a Chrome tab: https://code.earthengine.google.com/
Click the Earth Agent extension icon in your Chrome toolbar
Start chatting! The agent can see your Earth Engine environment and help with any GEE-related tasks

[!NOTE] The agent can only see the code in temp/new scripts in the Google Earth Engine Code Editor for now. You should save it once you think the script is ready.

Ask Mode vs Do Mode

The agent operates in two modes that you can switch between:

Ask Mode (Read-Only) 🔍

Analysis and guidance mode - the agent can observe but not modify:

✅ Read current code from the editor
✅ Search documentation (geeDocs)
✅ Take screenshots and inspect the page
✅ Read console output and map state
❌ Cannot insert or modify code
❌ Cannot execute code
❌ Cannot clear or reset the environment

Best for: Learning, understanding code, exploring datasets, getting explanations

Do Mode (Full Access) ⚡

Full execution mode - the agent can take action:

✅ All Ask Mode capabilities
✅ Write and edit code in the editor
✅ Execute code and monitor results
✅ Clear map, console, and inspector
✅ Complete end-to-end workflows

Best for: Building scripts, debugging, automated analysis, multi-step tasks

Tip: Start in Ask Mode when exploring new concepts, switch to Do Mode when ready to implement

Developer & Advanced Usage

🚀 MCP Server Support (Developer Version Only)

Earth Agent can work as a local Model Context Protocol (MCP) server, allowing you to control Google Earth Engine directly from Claude Code, Cursor, Zed, or other AI editors.

[!WARNING] Availability Note: The Chrome Web Store version of Earth Agent supports Cloud AI only (OpenAI, Anthropic, Google) due to browser security restrictions on localhost connections.

To use the MCP Server (or local Ollama models), you must install the Developer Version from GitHub.

How to Enable MCP Support

Install the Developer Version:
- Download earth-agent-extension.zip from GitHub Releases.
- Unzip and install via chrome://extensions/ > "Load unpacked".
Enable in Settings:
- Open Earth Agent settings.
- Find "External Integrations" and enable "MCP Server".

Configure your Editor:

Add the following to your MCP configuration (e.g., ~/.claude/settings.json):

hljs language-json

{
  "mcpServers": {
    "earth-agent": {
      "command": "npx",
      "args": ["-y", "earth-agent-mcp"]
    }
  }
}

👉 Full MCP Server Documentation

Usage Examples & Prompts

🌍 Getting Started

hljs language-arduino

Help me create a simple map showing NDVI for California using Landsat data

🛰️ Satellite Data Analysis

hljs language-sql

Find and display the latest Sentinel-2 image over New York City with less than 10% cloud cover

📊 Data Processing (⚡ Requires Do Mode)

hljs language-arduino

Calculate zonal statistics for land use categories within protected areas and export the results to Drive

🚀 Complex Workflow (⚡ Requires Do Mode)

hljs language-css

I want to analyze deforestation in the Brazilian Amazon from 2000 to 2023.
Please create a complete workflow that:
1. Loads appropriate satellite imagery
2. Calculates forest loss over time
3. Creates visualizations showing the changes
4. Generates statistics by state/region
5. Exports the results for further analysis

🔧 Debugging

hljs language-css

This code is running slowly, check my code for best practices and suggest improvements

Available Tools & Functions

Earth Agent provides a comprehensive set of tools available both in the Chrome Extension and via MCP.

Category	Agent Tool Name	MCP Tool Name	Description
Code Editing	`readCode`	`read_gee_code`	Read current code from editor
	`editCode`	`edit_gee_code`	Edit code using search & replace
	`writeCode`	`write_gee_code`	Overwrite entire editor content
	`undoEdit`	`undo_gee_edit`	Revert last edit
	`insertAtLine`	N/A	Insert text at specific line
Execution	`runCurrentCode`	`run_gee_code`	Run the script in editor
	`wait`	`wait`	Wait for execution/loading
Analysis	`getConsoleOutput`	`gee_console`	Read console logs & errors
	`getMapScreenPosition`	`gee_map_position`	Get map bounds & zoom
	`getInspectorOutput`	`gee_inspector`	Read inspector panel data
	`screenshot`	`gee_screenshot`	Take screenshot of interface
	`snapshot`	`gee_snapshot`	Get accessibility DOM tree
Knowledge	`geeDocs`	`gee_docs`	Search datasets & API docs
	`weather`	`weather`	Get weather data
	`dateTime`	`date_time`	Get current date/time
Control	`clearMapInspectorAndConsole`	`clear_gee`	Reset workspace
	`clickByRefId`	`click_element`	Click element by ID
	`clickAtScreenPosition`	`click_position`	Click at coordinates

[!NOTE] Z.AI currently does not support multimodal inputs yet, so screenshot analysis features are unavailable when using Z.AI models. For the best Earth Engine integration experience with visual analysis capabilities, use OpenAI, Anthropic, or Google providers.

Usage Tips

For Maximum Productivity

Use tool-compatible models when you need automated code execution
Be specific in your requests for better results
Ask for explanations to learn Earth Engine concepts
Use workspace management tools to keep your environment clean
Take screenshots to document your work and results

Common Workflows

Exploratory Analysis: Ask questions → Get code → Run and iterate
Data Processing: Define requirements → Generate pipeline → Execute and export
Visualization: Create map → Style layers → Add legends → Capture results
Debugging: Describe problem → Analyze code → Apply fixes → Test solution

Troubleshooting & Help

See Troubleshooting Guide for solutions to common issues (API keys, connectivity, tools, screenshots) and help resources.

Privacy and Security

We take your privacy seriously. Your API keys are stored securely in your browser using Chrome Sync and never share to our serever (even if we have one). Chat history and code remain local, it will sent to the LLM provider you choose. For full details on data handling and security practices, please read our Privacy Policy.

Roadmap

See roadmap.md for our planned features and future development.

Changelog

See CHANGELOG.md for a detailed history of changes.

Contributing

We welcome contributions! Please see our Contributing Guidelines for details on how to set up your development environment, build the project, and submit pull requests.

License

MIT

Thanks

Sundai Club
React
Vercel AI SDK

earth-agent-chrome-ext