Macuse - Give Your AI Superpowers on macOS

Hey Product Hunt,

I built Macuse because AI assistants are getting incredibly good at reasoning, but on macOS they still often hit a wall: they can answer questions, but they cannot reliably act across the apps where your work actually lives.

Macuse is a native macOS app that turns your Mac apps into local tools for AI assistants.

It runs as a local MCP server, connecting Claude, Codex, Cursor, Raycast, and any MCP-compatible client to your Mac. Your AI can manage Calendar events, read and draft Mail, work with Notes and Reminders, search Contacts, send Messages, and use Computer Use to click, type, scroll, and navigate real app interfaces.

A few things I cared about while building it:

• Local-first: your Mac app data is processed locally
• Permissioned: every connection requires approval and can be revoked
• Multi-client: one Macuse setup works across your AI tools
• Native integrations: Calendar, Mail, Notes, Reminders, Messages, Contacts, Shortcuts, Maps, and more
• Computer Use: control apps that do not have APIs, without taking over your active cursor/window

The goal is simple: make your AI assistant useful inside the Mac apps you already use every day.

I’d love feedback from Mac users, MCP builders, and anyone experimenting with AI agents on desktop workflows.

Love how it stays truly native on macOS instead of wrapping everything in an Electron shell. Tying Calendar, Mail, and Notes together through one MCP layer is the kind of plumbing I have been waiting for.

This looks great for the AI can reason but can't reach my apps problem. One thing I'm curious about: when Computer Use is driving clicks and typing in an app without an API, how do you handle it if it clicks the wrong thing in Mail or Messages before you can stop it? Is there some kind of review or undo step, or does it just fire at full speed once permission is granted?

connecting MCP clients to native mac apps is the missing layer for local workflows. right now claude can browse the web and run code but can't touch the calendar sitting right there on the same machine. the messages and mail access is where it gets interesting and also where it gets risky. what does the permission model look like? per-app grants, per-action approval, or one big trust decision at install? that choice basically determines whether people feel safe using it for anything real.

Hi Yuexun, the reassuring part for me is that it stays helpful without ever feeling like it might run off and do something behind my back. That sense of staying in the driver's seat matters a lot to me.

How does it actually handle permissions when an AI wants to send a message or move a calendar event, do you get a prompt each time or is there a way to set trusted rules upfront?