Build a Native Mac App with Claude Code

By the end of this talk you can start your own Mac app tonight — you'll know what to install, what's free, and the handful of decisions that make an app feel great.

Kevin Lenida

You describe. Claude codes. You decide.

Before you start

What you need — and what you can skip

You need (all free)

A Mac on the latest macOS
Xcode — Apple's app-building toolkit
Claude Code
A free Apple ID

You can skip

The $99/year Apple Developer account
Knowing Swift — or any code
Hiring a designer

Live build

The prompt that starts it all

Kevin

/grilling let's create a macOS todo app — fancy UI with liquid glass, a proper app icon, and MCP so AI agents can use it too.

Claude

Before I build anything — question 1: is this a real daily tool or a demo? My recommendation: build it for daily use…

Grilling skill: github.com/mattpocock/skills → productivity/grilling

Live build

Todo — the app we're building

The todo app — three-pane window with sidebar lists, tasks with due dates and tags, and a detail pane

01

The decisions that matter

Claude proposed most of this. My job was to say yes, no, or "convince me."

Decision · Framework

Native look, latest macOS

We went fully native — SwiftUI, Apple's own toolkit — targeting the latest macOS to get the Liquid Glass look.

Restraint wins: glass on the edges, solid where you read.

Watch out: new looks change fast. Make Claude check Apple's current docs instead of trusting its memory.

Decision · Framework

Same app, two skins

The todo app built natively — glass sidebar, macOS materials and depth

Native — feels like it belongs on your Mac

The same todo app mocked up as a generic web app — flat panels, web-style cards and buttons

Web-wrapped — a website in a window

Decision · Database

Boring wins

Your app's data needs a home. We chose SQLite — a 25-year-old database that lives in one file on your Mac. Proven, inspectable, no server.

Watch out: ask Claude "what's the boring, proven choice?" before accepting a database. Shiny new options cost you later.

Decision · App icon

Claude designed the logo

Claude designed our icon — described the app, got concepts back, picked one, refined it in plain language.

Mac icons are layered art, built in Apple's free Icon Composer — the system adds the shine and depth per layer.

Watch out: one strong shape over a simple background beats a busy scene. Icons live at tiny sizes.

Ten app icon variations Claude designed — different colors, motifs and styles, each labeled with a pick id

Decision · Apple Developer account

When free is enough

Free covers

Building & running on your own Mac
Daily use, forever — no expiry
Copying to another Mac you own

$99/year, only when

You give the app to other people
You want a clean, no-warnings download
You want the Mac App Store

Decision · Sharing

Share with a friend — still free

Send the project — they build it with free Xcode. Their Mac trusts what it builds: no warnings at all.
Send the app, off-internet — USB stick or file copy. No quarantine flag gets attached, so it just runs.
Send it over the internet — it arrives quarantined. They must approve it in Privacy & Security. Works, but looks sketchy.

Decision · App Store

Not every Mac app can go in

Fits the App Store

Minds its own business — touches only its own data
Opens files the user picks in a dialog
Normal internet use, standard permission prompts

Locked out

Controls or watches other apps
Roams your disk without asking
Installs system-level components
Wants to skip Apple's 15–30% cut

The App Store's one hard technical rule is the App Sandbox: every store app runs in a locked box that can only touch its own data, plus whatever the user explicitly grants through Apple's permission dialogs. Notes apps, todo apps, photo editors — fine. Window managers, keyboard remappers, backup tools, disk cleaners, developer tools that spawn programs — impossible, the sandbox forbids reaching across the system. Private Apple APIs are an instant rejection, and digital purchases must go through Apple's payment system at a 15–30% cut, which is why some apps skip the store on business grounds alone. Our todo app is the perfect edge case: even hosting the local MCP server is technically allowed, but publishing its endpoint to a fixed path any agent can find — cooperating with programs it didn't launch — is exactly what sandboxing exists to prevent. Rule of thumb: if the app minds its own business, the store is open; if its whole point is reaching into the rest of your Mac — or letting other programs reach in — it isn't. And that's the AI-era punchline: agent-controllable apps often live outside the store on purpose.

A modern app isn't just for humans — AI agents use it too.

Live moment

Let Claude grill you first

Every decision you saw tonight came out of an interview — Claude asked, one question at a time, and I answered.

The grilling skill makes that automatic: Claude questions your plan relentlessly before building anything.

You watched it happen at the start — that interview was this skill.

Get it: https://github.com/mattpocock/skills/blob/main/skills/productivity/grilling/SKILL.md

Recap

What to remember

Start free: a Mac, Xcode, Claude Code. The $99 waits until you ship to others.
Own the decisions — framework, database, icon. Say "convince me" to defaults.
Boring tech underneath, fancy only where people look.
Build for agents too — apps that AI can use are the new normal.

Your turn

Before we meet again

Install Xcode tonight (big download) — and Claude Code.

Pick one small Mac app you'd use every day. One window, one job.

Grill the idea first — make Claude interview you before it builds.

Let Claude build it, then actually use it for a week.

Bring it next time — working or broken. What you learned matters more than what you shipped.