Drive App Store Connect from any AI agent.

What you can do with it

Anything that's in App Store Connect's REST API — and a few things that aren't (chunked screenshot upload, "What's New" generated from git commits, AI-translated metadata across all locales, build attach-and-submit flow). Here's the surface area, grouped:

Build upload (which Apple ships through Transporter, not REST) is wrapped by upload_binary — agents can submit .ipa / .pkg files end-to-end. Only create_app (which Apple still gates to the App Store Connect web UI) returns an unsupported-feature note pointing the agent there.

The simulator-side tools let an agent do a full screenshot run end-to-end: boot the App-Store-required device matrix, install your build, launch it on each, screenshot in parallel, and save a folder of named PNGs to disk — all from one Claude Code session. Tap/swipe injection stays in the Mac app (it requires Accessibility permission tied to a specific bundle); everything else runs via the bundled sas-mcp shim.

Setup — three steps

The MCP server ships inside the Mac app bundle as sas-mcp. There's no separate CLI to install, no git clone, no brew tap, no PATH dance — and no credentials live anywhere except your Mac app's own storage.

1. Download & launch the Mac app

The app is a free download. The first launch walks you through pasting in your App Store Connect API key — Key ID, Issuer ID, and .p8 file. From then on the embedded MCP server can talk to Apple on your behalf.

2. Keep the app open in the background

The bundled sas-mcp binary is a thin stdio shim — your agent client spawns it as a subprocess, and it proxies JSON-RPC calls to the Mac app's local server. As long as Simple App Shipper is running, every call is logged in the in-app Agent Activity Feed so you can audit (or pause) whatever your agent is doing in real time.

3. Add one block to your agent config

Pick your client. Snippets are below — same shim, three config formats.

Connect your client

Pick whichever you use day-to-day. All of them speak MCP.

Claude Code — ~/.claude/mcp_settings.json

{
  "mcpServers": {
    "simple-app-shipper": {
      "command": "/Applications/Simple App Shipper.app/Contents/MacOS/sas-mcp"
    }
  }
}

Cursor — ~/.cursor/mcp.json

{
  "mcpServers": {
    "simple-app-shipper": {
      "command": "/Applications/Simple App Shipper.app/Contents/MacOS/sas-mcp"
    }
  }
}

Codex — ~/.codex/config.toml

[mcp_servers.simple-app-shipper]
command = "/Applications/Simple App Shipper.app/Contents/MacOS/sas-mcp"

No credentials in the snippet, no env vars, no tokens — the shim asks the running Mac app, which holds your ASC API key in its own secure storage. If you moved the app somewhere other than /Applications, point command at <your path>/Simple App Shipper.app/Contents/MacOS/sas-mcp instead.

Real example prompts

The agent calls the right tools automatically. You don't memorize tool names — just say what you want.

Prep a release

› Bump my app's version to 2.4.0, write release notes from the
   last 14 commits, AI-translate "What's New" into every locale,
   attach build 47, and stage it for review (don't submit).

Audit a release

› For my app, list versions in review or rejected. For each,
   show the rejection reason, the contact info on the review
   detail, and any failed pre-submit checks.

A/B test an icon

› Create a PPO experiment named "icon-v3-test" at 80% traffic
   with my candidate icon vs the current default. Print the
   App Store Connect analytics URL when done.

Localize an entire listing

› Read my en-US version metadata, then translate the description,
   promotional text, what's new, and keywords into all 28 locales
   on this version. Save as drafts. Don't submit.

Generate App Store screenshots end-to-end

› List the simulators. Boot iPhone 17 Pro Max, iPhone 11 Pro Max,
   iPhone 8 Plus, iPad Pro 13", and iPad Pro 12.9". Install
   ~/Build/MyApp.app on all five. Launch com.example.MyApp on each.
   Wait 3 seconds, then screenshot all five in parallel into
   ~/Desktop/AppStoreShots/screen-1/. Open simpleappshipper://onboarding-step-2
   on all five, wait 2 seconds, screenshot into screen-2/. Repeat for
   step 3.

Safeguards

• Submit for Review still requires a human password set at app onboarding. The agent prepares everything and stops. You press Submit.
• Destructive deletes (apps, versions, builds, IAPs, screenshots) require the same password.
• Activity feed in the Mac app logs every tool call with method, params, status, duration. Pause or audit at any time.
• Localhost-only. The embedded server in the Mac app binds 127.0.0.1, never the LAN.
• Your credentials. Your ASC API key never leaves disk; the agent calls Apple's API through the server, not directly.

Get started

Download is free. The Mac app, the MCP server, and the agent surface come in one bundle — no separate CLI, no GitHub clone, no build step. AI features (image generation, vision analysis, screen-to-HTML) run on credits; Pro is $7.99/mo for 30k credits if you want them.