Edgebox, developed by Bigppwong, is a specialized MCP server tool designed to provide Large Language Model agents with extensive "Computer Use" capabilities. Built using Electron and TypeScript, it operates as a local desktop sandbox, offering a complete desktop environment and secure code execution. Distributed as binary releases across Windows, macOS, and Linux, Edgebox allows AI agents to interact programmatically with desktop applications. It connects to LLM agents via the Model Context Protocol, enhancing their ability to perform complex real-world tasks.

What does Edgebox connect to?

Edgebox primarily connects to Large Language Model agents through the Model Context Protocol (MCP). It functions as a local desktop sandbox for these AI agents, giving them a full desktop environment. This setup allows LLMs to execute code and interact with the operating system as if they were directly operating a computer. Its core purpose is to extend the capabilities of AI agents beyond text-based interactions, enabling them to manipulate graphical user interfaces.

Core features and tools

Edgebox exposes a range of callable tools for AI agents. Core Tools, always available in CLI Mode, include shell_run, shell_run_background, container_list, container_create, container_stop, container_restart, and container_delete. When GUI Tools are enabled, Desktop Tools become available, such as desktop_screenshot, desktop_mouse_click, desktop_keyboard_type, and desktop_clipboard_read. These tools allow granular control over both the command line and desktop environment.

Who is it for?

Edgebox is designed for developers and users who want to empower Large Language Models with "Computer Use" capabilities. Its functionality allows AI agents to interact directly with a full desktop environment, performing actions like automating desktop applications or running code securely. This tool helps bridge the gap between AI reasoning and practical computer interaction. Skip this if your AI agent tasks do not require direct desktop or containerized code execution.

How to get started

No credentials are required for the MCP server itself. Authentication primarily involves client-side configuration and the use of environment variables for sensitive values, with session isolation managed by passing different x-session-id headers via streamable HTTP.

Use cases

• Automating interactions with pre-installed desktop applications like Google Chrome and VS Code using programmatic mouse and keyboard control.
• Executing AI-generated code securely within isolated Docker containers and interacting with the Linux shell environment.
• Managing multiple, isolated sandbox sessions concurrently for different AI agent tasks or testing purposes.

Edgebox provides a powerful local sandbox for AI agents, enabling them to interact with a desktop environment and execute code securely. Its multi-platform binary releases and support for streamable HTTP make it accessible for various setups. The tool extends LLM capabilities to automate complex desktop workflows and manage containerized operations effectively. Try automating a desktop application like Google Chrome using the desktop_mouse_click tool.