"Actual Budget" matching MCP servers:

macOS automation for AI assistants: 17 typed tools for window management, application menus, keyboard input, clipboard, browser tabs, screenshots, displays and Apple Shortcuts, driven through AppleScript and JXA. Menu clicks are self-correcting — a wrong item returns the actual menu contents so the model can retry with the right one.

An MCP server that lets Claude operate your real computer by moving the actual mouse, clicking, typing, and reading the actual screen, working with your own logged-in sessions in any application.

Publishes two MCP servers that let an agent drive a real machine: one automates a Playwright browser (navigate, snapshot element refs, click, type, screenshot, download files), and one controls the macOS desktop by taking screenshots and moving the actual mouse and keyboard across any native app. Useful for web tasks with element-level precision and for OS-level work like launching apps, using menus, and keyboard shortcuts.

Exposes guarded system-automation tools to MCP clients, enabling allowlisted shell execution, content-aware file organization, rsync backups, VAAPI transcoding, disk audits, application inspection, and delegation of coding tasks to CLI agents. Every action runs under dry-run defaults, path containment, deny-globs, a kill switch, budget caps, and SQLite audit logging.

Enables agents to capture the actual desktop screen as base64 PNG images, with optional region, monitor, display, backend, and path targeting. Also exposes a tool for listing which native capture backends are available in the current environment.

Enables desktop AI clients such as Claude Desktop, Cursor and ChatGPT to reach Wave Mobile Money on an Android phone over local Wi-Fi, executing official USSD requests and reading SMS or notifications to check balances and send transfers. Bundles a Family Officer doctrine that pre-evaluates each transfer's budget impact, safeguards a minimum emergency reserve, and shields users from fraudulent Wave messages.