Clawd Cursor

What it does

Clawd Cursor provides an OS-level desktop automation server, effectively giving AI models the ability to interact with a real computer. It allows any AI model – including Claude, GPT, Gemini, and Llama – to perform actions like moving the mouse, capturing screenshots, and automating tasks through a tool-calling interface. The system is designed to be model-agnostic and can even function with locally run models.

When to use it

• Automated data entry: Have an AI agent fill out online forms or spreadsheets by controlling keyboard input and mouse clicks.
• Web scraping & research: Automate the process of navigating websites, extracting information, and saving results.
• Software testing: Use an AI to perform repetitive software testing tasks, simulating user interactions.
• Content creation workflows: Automate parts of content creation processes like image editing or video rendering by controlling desktop applications.

Key capabilities

• Model-agnostic compatibility (Claude, GPT, Gemini, Llama)
• Works with locally run AI models
• Mouse movement and dragging (moveMouse, dragMouse)
• Screenshot capture (captureScreen)
• macOS native host app for simplified permission management
• Localhost IPC architecture for enhanced security

Example prompts

• "Move the mouse to coordinates 100, 200 and click."
• "Capture a screenshot of the current window and save it as 'screenshot.png'."
• “Open Chrome, navigate to clawdcursor.com, and scroll down three times.”

Tips & gotchas

• On macOS, you'll need to grant Accessibility and Screen Recording permissions to the Clawd Cursor Host App for full functionality. This is a one-time setup.
• The system uses a token-based authentication via ~/.clawdcursor/host-token for secure communication; ensure this file has appropriate permissions (0600).