Generate App Store Screenshots With Claude Code

You can generate a full App Store screenshot set from inside Claude Code by connecting Screenies' remote MCP server: run one claude mcp add command, sign in through the browser once, then ask the agent to upload your screens and generate the set. This guide walks that setup and the prompt-driven workflow it opens, then draws the line between it and the local Claude Code skills that also rank for this search.

Search this exact phrase and the results split into two camps that behave very differently, so it's worth sorting them before you install anything. Most of the top results are skills: markdown-and-script bundles like ParthJadhav's app-store-screenshots, which scaffolds a Next.js editor into your repo, or Adam Lyttle's ASO skill, which reads your codebase and assembles frames locally, often leaning on fastlane for the device mockups. The other camp is remote MCP servers, where the generation runs on someone's hosted service and the agent just calls tools over a connection: AppLaunchFlow and AppScreenshotStudio both offer that, and Screenies sits here too. The two camps trade off in a clean way. A local skill gives you the files and full control but leaves the design, the spec upkeep, and any build errors on your machine; a hosted server hands off all of that and asks only that you sign in.

Skills that scaffold versus servers that generate

The practical difference shows up the first time something breaks. A skill that scaffolds a Next.js editor means you now own a small web app: its dependencies, its render step, and whatever the image model decided your captions should say, all sitting in your working tree. That's the right trade if you want to hand-tune every pixel and keep the source. A hosted MCP server keeps none of that on your side, so there's no local render to debug and no generated code to review, at the cost of the artwork living on the service rather than in your repo. Screenies' MCP server is the hosted kind: the same generation that runs in the web dashboard, exposed to your agent as tools. If you'd rather read the tool list and connect from a client page, the per-client setup pages cover Claude Code, Claude.ai, and Cursor.

Connecting the server to Claude Code

Setup is a single command you paste into your terminal:

claude mcp add --transport http screenies https://screenies.app/mcp

Then run /mcp inside a Claude Code session, pick screenies, and the agent opens a browser window where you sign in to Screenies and approve the connection. That's OAuth 2.1 doing the work: your agent never sees your password, the tools are scoped to your own workspaces and apps, and you can revoke access whenever you like. There are no API keys to paste or rotate. Any other client that speaks remote MCP takes the same URL, so the Cursor and Claude.ai flows differ only in where you enter it. For the editor version, the Cursor screenshot workflow covers the mcp.json entry and which tool calls to approve. Using Codex instead? See generating App Store screenshots with Codex. If you're deciding what else to connect alongside it, the best MCP servers for Claude Code list covers the Sentry, Supabase, and Vercel servers that pair well with a release workflow.

For the server implementation, our remote MCP OAuth guide explains discovery paths, token verification, and workspace access checks.

Generating a set from prompts

Once connected, the whole set runs from the chat. Tell the agent what you're launching and it walks the tools in order: list_apps to find or place your app, request_screenshot_upload to get signed URLs for your raw captures, then generate_screenshot_set to start the run against your uploaded screens and a short description of what the app does and why it's worth downloading. Generation is asynchronous, so the agent polls get_batch_status until the slides finish, and each completed slide comes back with a download URL. If a slide fails, regenerate_failed retries just that one, and failed generations refund their credit automatically. A typical prompt is as plain as "upload these five screens, generate a set for a habit tracker aimed at people who keep quitting," and the agent handles the tool calls from there.

One property matters more than convenience here: every slide shows your real app. A screen sits in a device frame exactly as captured or is rendered into a photographic scene from your capture, and nothing is invented. That is what keeps a listing clear of the failure mode developers keep hitting, where a generator invents UI and the submission gets rejected for not matching the actual app under guideline 2.3.3. What the agent controls is the pitch around your screens: slide order, the headline on each slide, and a set-wide palette pulled from your UI. What it can't do is invent functionality you don't have.

Localizing and editing without leaving the terminal

Revisions are tools too, so you don't drop back into a design app to fix a word. edit_headline rewrites the caption on one slide, reorder_set changes the sequence and drags any localized copies along with it, and headline edits after generation don't spend credits. When the English set is right, localize_set translates it into other App Store locales; because it reuses the artwork instead of calling the image model again, adding a fifth language costs 0.1 credits per screenshot rather than a fresh generation. You can check get_credit_balance before a big run if you're watching the meter. Everything you started in Claude Code stays on the same account, so you can open the dashboard later and the app, its set, and its credits are all there.

Open the MCP connection guide to connect your agent to your Screenies workspace.

Cost, and what the connection doesn't change

The MCP server is included with every Screenies account, and generation started from your agent spends the same credits as generation started in the dashboard: one credit per generated phone slide (iPhone, iPhone Duo outer, or Android phone), two credits per tablet-sized slide (iPad, iPhone Duo inner, or Android tablet), free headline edits, and localization at 0.1 credits per screenshot, per locale. You can request phone or tablet output in portrait or landscape, with matching source screenshots and one device and orientation per batch. The pricing page lists the same one-time credit packs for both workflows.