Hawa Computer Use

Hawa Computer Use is a cross-platform computer control server for Hawa Code Agent, allowing AI to automate operations on your computer through screenshots, mouse, and keyboard. It supports Windows, macOS, and Linux.

Installation

Install with:

npm install -g @dahawa/hawa-computer-use

Use as an MCP Server (Recommended)

Add the following configuration in Hawa Code or other MCP-compatible clients:

{
  "mcpServers": {
    "computer-use": {
      "command": "npx",
      "args": ["-y", "@dahawa/hawa-computer-use"]
    }
  }
}

Provided MCP Tools

`computer`

Interact with the computer through mouse and keyboard, and capture screenshots.

Supported actions (action):

Action	Description
`get_screenshot`	Capture a screenshot and return an image with element annotations and an element list
`mouse_move`	Move the mouse to the specified coordinates
`left_click`	Left-click (optionally move to specified coordinates first)
`right_click`	Right-click
`middle_click`	Middle-click
`double_click`	Double-click
`left_click_drag`	Drag to the specified coordinates
`scroll`	Scroll the wheel, supports `up`/`down`/`left`/`right`, and can specify pixels such as `down:500`
`key`	Press a keyboard key or key combination, such as `ctrl+c` or `alt+tab`
`type`	Type text (automatically handles non-ASCII characters such as Chinese)
`get_cursor_position`	Get the current mouse coordinates
`click_element`	Click by element ID (recommended first choice; more accurate than coordinate clicking)

System Dependencies

Different platforms require the following optional dependencies to unlock full functionality:

Windows

Works directly on Windows.

MacOS

AX detection: Grant accessibility permission to the host application (such as Terminal or Hawa Code) in “System Settings → Privacy & Security → Accessibility”
OCR detection (optional): Install Xcode Command Line Tools
xcode-select --install

Linux

AT-SPI detection:

# Debian / Ubuntu
sudo apt install python3-pyatspi

# Fedora / RHEL
sudo yum install python3-pyatspi

OCR:

# Debian / Ubuntu
sudo apt install tesseract-ocr tesseract-ocr-chi-sim

# Fedora / RHEL
sudo dnf install tesseract tesseract-langpack-chi_sim

Screenshot tools (install at least one): gnome-screenshot, scrot, or ImageMagick import

Environment Variables

Variable	Description	Default
`MCP_TRANSPORT`	Transport protocol, `stdio` or `http`	`stdio`
`PORT`	Listening port in HTTP mode	`3000`
`COMPUTER_MCP_DISABLE_A11Y`	Set to `1` to disable accessibility detection and use coordinate mode only	-
`COMPUTER_MCP_DISABLE_OCR`	Set to `1` to disable OCR fallback and use accessibility detection only	-

Hexo

computer use