Skip to main content

App Store Screenshot MCP Servers: Which One Designs the Set

Nine servers turn up for this search. Four of them capture pixels or upload files and never design anything. These are the five that build the frames, compared from their own docs.

Renders come back inside the chatClaude Code, Cursor, Windsurf, CodexSeven sizes, App Store and Play
App Store screenshot generated over MCP for a running app
App Store screenshot generated over MCP for a habit tracker
App Store screenshot generated over MCP for a budgeting app
App Store screenshot generated over MCP for a podcast app
App Store screenshot generated over MCP for a meditation app
The real shortlist

Five servers design screenshots. Four only move pixels.

An MCP server with “screenshot” in the name might drive a simulator, lay out a marketing frame, or upload a finished file to Apple. Only the middle one is a screenshot tool in the sense you mean. These are those five.

MCP servers that design App Store screenshots, compared on where the design comes from, store sizes, localization, platform and access
ServerDesign fromStore sizesLocalizationRuns onAccess
@appscreenshotstudio/mcpoursPrompt + your repo7, App Store + PlayAll 50 App Store languages, in the builderAny OSGrowth or Pro plan
appshots-mcpTypst templates2 required sizes39 ASO localesmacOSApache-2.0 or MIT, free
ButterKitPromptnot published50 App Store languagesmacOSFree, $99 to unwatermark
AppLaunchFlowCanvas you editnot publishedAny languageAny OSFree tier includes MCP
appscreen-mcpEditor you script6, plus customAny languageNot statedMIT, free

Design from is the column that matters most and the one nobody publishes side by side. Two of these servers build a layout from a description. Three hand you a template or a canvas and expect you to drive it, which is a different product wearing the same word.

Tool count is not a column, on purpose. It measures how a project split up its API, not what it can do. The two publish servers further down settle it: one exposes 91 tools, the other reaches 923 Apple endpoints through 2. Every tool’s schema also loads into your agent’s context on every turn, so a big number is a running cost as often as a capability. The verified counts sit on each server’s card below.

Verified against each project’s own README, docs or npm entry on 2026-08-13. “Not published” means the project documents the feature without documenting the specifics, which is not the same as a no. StoreScreens was a candidate and is absent because its site returned an HTTP 403, so there was nothing to check.

Pick by need

Which one to install

Every claim behind these is in the table above, with a link to the source on each card below.

You want the brief pulled out of your codebase, and you build on Windows or Linux

@appscreenshotstudio/mcp

You need Play Store sizes as well as App Store sizes

@appscreenshotstudio/mcp

You ship a listing in 40 or more languages

ButterKit, or appshots-mcp

You want free and open source, and Xcode is already installed

appshots-mcp

You want nothing to install and a free tier

AppLaunchFlow

Your unreleased UI must never leave the laptop

appscreen-mcp
Server by server

What each one is actually good at

Including the thing it does better than we do. A table where our row wins everything would not be worth reading.

@appscreenshotstudio/mcp

Node, any MCP client· 9 tools

Reads your repo for the brief, designs a full set from it, takes your app UI into the frames, renders at exact store sizes.

Where we lose: Localization is where we lose. ButterKit covers all 50 App Store languages and appshots-mcp resolves 39 ASO locales with fallback chains. We translate 15, and that runs in the builder rather than as an MCP tool.

Setup guide

appshots-mcp

macOS, Xcode, Typst· 24 tools

Rust server that captures clean simulator frames, then composes them with Typst. Aligns captions to ASO keywords.

Where it beats us: The only server here that both captures and composes, it is free and open source, and Typst gives it native RTL and CJK layout.

Check the source

ButterKit

ButterKit.app v1.8+, macOS· 40 tools (their overview says 42)

Designs from a prompt, localizes, and pushes the result to App Store Connect without leaving the agent.

Where it beats us: The only server here that composes and publishes, so it is the shortest path from prompt to a live product page. It also covers all 50 App Store languages.

Check the source

AppLaunchFlow

Remote HTTP, no install· tool count not published

Hosted canvas driven over a remote MCP endpoint. Edit headlines, swap colours, translate, export.

Where it beats us: MCP access is on every plan including the free one, where ours needs a paid plan. It is also a remote endpoint, so there is nothing to install.

Check the source

appscreen-mcp

Local app, Playwright, Chromium· tool count not published

Drives a local browser-based screenshot editor through a scripted API rather than by scraping the DOM.

Where it beats us: Runs entirely on your machine with no account and no upload, so your unreleased UI never leaves the laptop.

Check the source
The other four

The most literally named server produces a file Apple will not accept

mcp-ios-simulator-screenshot defaults to resize: true with a max_width of 640 pixels. App Store Connect will not accept an iPhone screenshot narrower than 1242.

That is not a bug. Capture servers downscale on purpose, because they exist to let an agent see your app cheaply, and a full-resolution frame burns tokens for no benefit. The default is right for the job it was built for and wrong for yours.

The deeper problem survives even at full resolution: a raw simulator frame has no headline, no background and no device frame. It is a picture of your app, not a marketing asset. That is why these four sit here rather than in the table, and why a fully hands-off pipeline means two servers wired into the same agent.

Capture

Gets raw pixels out of a running simulator. These are built so an agent can see your app, which is not the same as building a file Apple will accept.

Drives the simulator: tap, type, swipe, inspect the accessibility tree, screenshot, record video.

macOS, Xcode, Facebook IDB

mcp-ios-simulator-screenshot· 1 tool (get_screenshot)

Captures the booted simulator screen to a directory. One tool, no configuration to speak of.

macOS, Node 18+

Publish

Pushes finished images to App Store Connect. Every server in this column uploads files it did not create and cannot design.

app-publish-mcp· 91 tools

Uploads finished images to App Store Connect and manages listings, releases, reviews and submissions.

App Store Connect API key and .p8

appstore-connect-mcp· 2 tools (923 endpoints behind them)

Exposes Apple's whole API through a search-and-execute pair instead of one tool per endpoint.

App Store Connect API key and .p8

Ours, concretely

Four calls from your repo to store-ready PNGs

Add this to your MCP client config. Claude Code, Cursor, Windsurf and Codex all read the same shape.

{
  "mcpServers": {
    "appscreenshotstudio": {
      "command": "npx",
      "args": ["-y", "@appscreenshotstudio/mcp"],
      "env": { "APPSCREENSHOTSTUDIO_API_KEY": "ass_live_..." }
    }
  }
}
1
prepare-screenshot-brief

Returns a category playbook and the exact list of things to dig out of this repo. Free, and it runs before anything gets designed.

2
generate-screenshots

Designs the set from that brief and returns a project with empty device frames.

3
upload-screenshots

Drops your real app UI into those frames from local file paths.

4
render-screenshots

Exports PNGs at store dimensions. Free, so iterate as much as you want.

The constraints, stated plainly. We design and render. We do not capture from a simulator and we do not upload to App Store Connect, so a hands-off pipeline pairs this with one of the four above. Localization runs in the builder rather than as an MCP tool, and covers all 50 App Store languages. MCP access needs a Growth or Pro plan, because it runs on the same API keys as the REST API.

Output

What comes back from render-screenshots

Real frames from a set designed over MCP, at the dimensions App Store Connect accepts. No mockup, no illustration.

Rendered App Store screenshot, frame 1 of a fitness app set
Rendered App Store screenshot, frame 2 of a fitness app set
Rendered App Store screenshot, frame 3 of a fitness app set
Rendered App Store screenshot, frame 4 of a fitness app set

Seven device sizes ship in the server, from iPhone 16 Pro Max at 1260 x 2736 down to Apple Watch Ultra 2 at 410 x 502, including three Play Store sizes. Browse more in the builder.

Questions about MCP screenshots

Does Screen Studio have an MCP server?

No. Screen Studio is a macOS screen recorder for demo videos and tutorials, and as of August 2026 its site documents no MCP server or Model Context Protocol integration. It is a common search because the name sounds adjacent, but it does not produce App Store screenshots at all.

Which MCP server should I install for App Store screenshots?

Five of the nine design anything, so the choice is among those. If you want the brief pulled out of your codebase and you are not on a Mac, AppScreenshotStudio fits. If you want the translation pass itself driven by an agent rather than run in a builder, ButterKit scripts it end to end. If you want free and open source and Xcode is already installed, appshots-mcp. Capture and publish servers are a second install rather than an alternative, because neither designs anything.

Which App Store screenshot MCP server has the best localization?

ButterKit and AppScreenshotStudio both cover all 50 App Store languages, with appshots-mcp close behind at 39 ASO locales including fallback chains such as es-MX to es-ES. The real split is where localization happens: ours runs in the builder rather than as an MCP tool, so an agent cannot trigger a translation pass unattended. ButterKit is the pick if you want that step scripted end to end.

Can one MCP server do the whole pipeline?

Close to it, but not quite. ButterKit composes and publishes, and appshots-mcp captures and composes. Nothing verified here does all three of capture, compose and publish, so a fully hands-off pipeline currently means two servers wired into the same agent.

Why can I not just use an iOS simulator screenshot MCP?

You can capture with one, but the output is not submittable. Simulator capture servers are built to feed an agent vision, so they downscale by default: mcp-ios-simulator-screenshot ships with a 640 pixel maximum width. App Store Connect will not accept an iPhone screenshot below 1242 pixels wide, and a raw simulator frame carries no headline, background or device frame either.

Is fastlane an MCP server?

No. Fastlane is a Ruby toolchain, and Snapshot and Deliver run as lanes from a Fastfile rather than as tools an agent discovers. It solves a similar problem from the CI side rather than the agent side, and the two combine well: capture with Snapshot, compose over MCP, upload with Deliver.

Do these work in Cursor and Codex, or only Claude Code?

Any client that speaks the Model Context Protocol, which includes Claude Code, Cursor, Windsurf, Codex and Zed. The capture servers are the exception to portability, since they need macOS with Xcode installed no matter which client is calling them.

What does the AppScreenshotStudio MCP server cost?

It needs a Growth or Pro plan, because MCP calls run through the same API keys the REST API uses and the Starter plan does not include API access. Rendering never costs credits once a design exists, so iterating on layout is cheaper than the first generation.

How current is this comparison?

Every capability on this page was read from the project's own README, docs or npm entry on 2026-08-13. Where a project does not publish a fact, this page says so instead of estimating it, which is why several cells read "not published" rather than carrying a number. MCP servers move quickly, so check the linked source before making a decision that matters.

Point your agent at it

Install the MCP server, then ask Claude Code or Cursor for a set. It reads your repo for the brief, seven device sizes, rendering always free.