An open-source MCP server for Claude Code, Codex CLI, Gemini CLI and Cursor. It validates the idea with real App Store data, builds the SwiftUI app, sets up the backend and the store, and stops at every step that matters so you stay in control.

claude mcp add appfactory -s user -- uvx appfactory@latest

A real, sped-up agent session: AppFactory scores water-tracking niches with live App Store data

A real session, sped up. That first step needs no accounts. From there, the same agent can take the idea all the way to a TestFlight build.

What it does

flowchart LR
    A[💡 Idea<br/>real App Store data] --> B[🎨 Design<br/>competitor research]
    B --> C[📱 SwiftUI app<br/>from a tested template]
    C --> D[☁️ Backend<br/>Supabase]
    D --> E[💳 Store<br/>IAPs + RevenueCat]
    E --> F[🌍 Localize<br/>+ ASO + screenshots]
    F --> G[🚀 TestFlight]
    G -.you submit.-> H[App Store]

Stage	What the agent does with AppFactory
Idea	Harvests rising and chart-proven apps, scores niches by search demand, competition and staleness, estimates competitor revenue, checks the name is free
Design	Downloads category leaders' screenshots and icons, writes a differentiated design brief, designs every screen
App	Scaffolds a SwiftUI app (iOS 17.5+, iPhone) with onboarding, paywalls, settings, analytics hooks and tests
Backend	Supabase project, auth, edge functions, server-side usage limits, optional AI proxy (keys never ship in the app)
Store	App Store Connect app, subscriptions with trials and local prices, RevenueCat offerings, StoreKit config kept in sync
Launch assets	Localized metadata, ASO keywords, legal pages, branded screenshots, preview videos, custom product pages
Ship	Signing, archive, TestFlight upload. You press "Submit for Review".

Every stage has a gate: the agent can't mark it done until checks pass (the build compiles, the store matches the spec, screenshots follow the rules...). When something needs you, the run stops with a clear NEEDS_HUMAN.md.

Quickstart

Requirements: a Mac and uv (brew install uv). Xcode and the other tools are only needed for the services you turn on.

1. Add it to your agent

Agent	Command
Claude Code	`claude mcp add appfactory -s user -- uvx appfactory@latest`
Codex CLI	`codex mcp add appfactory -- uvx appfactory@latest`
Gemini CLI	`gemini mcp add appfactory uvx appfactory@latest`
Cursor	`~/.cursor/mcp.json`: `{"mcpServers": {"appfactory": {"command": "uvx", "args": ["appfactory@latest"]}}}`

2. Say "Set up AppFactory"

The agent asks which services you want, one by one, and tells you exactly which keys are missing. Keys are environment variables on the MCP entry, like any other MCP server. They never go through the chat:

claude mcp add appfactory -s user \
  -e APPFACTORY_ASC_KEY_ID=ABC123DEFG \
  -e APPFACTORY_ASC_ISSUER_ID=00000000-0000-0000-0000-000000000000 \
  -e APPFACTORY_ASC_KEY_FILEPATH=~/keys/AuthKey_ABC123DEFG.p8 \
  -e APPFACTORY_TEAM_ID=TEAMID1234 \
  -- uvx appfactory@latest

Every variable is listed in docs/SETUP.md.

3. Run it

Run AppFactory

Before any work the agent asks you about every optional part of the app: free trial or not, hard paywall, discounted offer, onboarding quiz, mascot, languages, screenshots, preview video, analytics, ratings, Sign in with Apple... Anything you turn off is skipped, in the app code too.

Choose what you use

Everything except research is optional. A disabled service's tools do nothing and its stages are skipped, and the generated app doesn't include its SDK.

Service	What it enables	Needs
research	Idea harvesting and validation, ASO research	nothing
apple	App Store Connect: apps, subscriptions, metadata, TestFlight	ASC API key, `asc` CLI
xcode	Local builds, simulator, archives	Xcode, `xcodegen`
supabase	Backend: database, auth, edge functions, usage limits	access token, `supabase`, `deno`
revenuecat	Subscriptions and entitlements (without it: StoreKit 2 only)	RevenueCat secret key
ai	AI features through a server-side proxy (needs supabase)	provider keys, optional
firebase	Firebase Analytics	`firebase` CLI
github	A private repo and issue workflow per app	`gh`
design	Screen design through the claude-design MCP (without it: design locally)	claude-design MCP
maestro	End-to-end UI tests on the simulator (needs xcode)	Maestro, Java 17+
lottie	Lottie animations	`node`

Safety

Agents make mistakes and read untrusted web content, so AppFactory assumes they will.

Human approval for anything irreversible. Store writes, backend deploys, database changes, GitHub pushes, uploads and signing don't run when the agent calls them. They wait until you run appfactory approve <id> in your own terminal. The agent can't approve on your behalf.
You submit. App Store review submission is always done by you.
Dry run by default for deploys, repo creation and uploads. Destructive SQL always needs approval.
Secrets stay on your machine and are masked in every tool result.
Third-party text is data. App Store listings and reviews are labelled so the agent doesn't follow instructions hidden in them.

Threat model and limits: SECURITY.md.

FAQ

Does it need Claude? No. It is a standard MCP server. The run playbook is served by the server itself (tool playbook(), prompt run, resource appfactory://playbook), so any MCP-capable agent can follow it.

Why a Mac? The app is built and tested with Xcode on your machine. Research works anywhere.

Is my data sent anywhere? Only to the services you turn on, with your own keys. There is no AppFactory server and no telemetry.

Can I change the rules? Yes. The defaults (subscription with a free trial, hard paywall plus a discounted offer, a quiz onboarding, 6 app and 8 store languages) are recommendations. You answer each one before a run.

How do I update? @latest in the MCP command picks up new releases when your agent restarts.