AppMarquee
A local-first MCP server for creating and managing App Store and Google Play screenshot decks, letting agents upload, review, localize, and export store-ready images without accessing source code.
README
<div align="center"> <img src="plugins/appmarquee/assets/icon.svg" width="84" alt="AppMarquee logo" /> <h1>AppMarquee</h1> <p><strong>Your store-screenshot studio—built for people and agents.</strong></p> <p>Create, review, localize, and export App Store and Google Play screenshot decks from one local workspace. Give agents the screenshots, not your source code.</p>
English · Türkçe </div>

The short version
Most screenshot generators begin as code inside each app repository. AppMarquee is one reusable program instead:
- Create a project in AppMarquee.
- Upload real product screenshots.
- Ask Codex, Claude, or another MCP agent to build the story.
- Review visually and export every required store size.
The agent never needs the app’s source directory. Your project files and screenshots remain local.
What ships
| Studio | Agent bridge | Output |
|---|---|---|
| Multi-project dashboard + bulk import | Codex plugin + MCP server | App Store iPhone and iPad sets |
| Connected or isolated canvas | Claude Code and generic MCP setup | Google Play phone and tablet sets |
| Drag, resize, rotate, reorder | Visual screenshot reading | 1024×500 feature graphics |
| Themes and localized copy | Revision-safe targeted updates | Language-selectable PNG ZIP bundles |

Quick start
Requires Bun 1.1+ and Node.js 20+.
git clone https://github.com/furkangonel/AppMarquee.git
cd AppMarquee
bun install
bun run mcp:build
bun run dev
Open http://127.0.0.1:3000.
Connect Codex
bun run mcp:install:codex
Connect Claude Code
bun run mcp:install:claude
Connect Claude Desktop
bun run mcp:install:claude-desktop
Completely quit and reopen Claude Desktop after installation. See the complete Claude Desktop guide for manual configuration, verification, removal, troubleshooting, and Desktop Extension status.
Restart the agent, then ask:
Check AppMarquee status. Inspect the screenshots in Focus Garden and create
a six-screen iPhone deck in English and Turkish. Validate it when finished.
That is enough. You can launch the agent from another folder; the generated MCP configuration uses an absolute path.
For plugin marketplace installation, generic MCP JSON, verification, and removal instructions, see the complete installation guide.
Designed for a real launch workflow
01 — Bring the product proof
Select or drop several simulator, emulator, or device captures in one pass. AppMarquee sorts them by filename, fills empty screens, and adds screens up to the active store limit. Assets use content-addressed filenames and stay inside that local project.
02 — Let the agent shape the narrative
The agent can view uploaded screenshots, write benefit-led copy, choose layouts, localize text, and replace only the requested device deck. Revision checks prevent stale agent writes from silently replacing newer human edits.
03 — Keep the final visual decision human
Reload agent changes in the editor, adjust anything visually, choose exactly which project languages to download, and export exact PNG bundles. AppMarquee supports iPhone, iPad, Android phone, 7-inch tablet, 10-inch tablet, and Google Play feature graphics.
MCP tool surface
| Tool | What it lets an agent do |
|---|---|
appmarquee_status |
Confirm that the local studio is available |
appmarquee_list_projects |
Discover local projects without a directory path |
appmarquee_create_project |
Start a new launch workspace |
appmarquee_get_project |
Read project state and its current revision |
appmarquee_list_assets |
Find uploaded product screenshots |
appmarquee_read_asset |
View an uploaded screenshot as image content |
appmarquee_update_device_deck |
Replace one device deck without touching the others |
appmarquee_update_settings |
Change name, theme, canvas mode, or locales |
appmarquee_validate_project |
Check structure and Apple/Google screen limits |
appmarquee_project_link |
Return the exact local visual-review URL |
Why it stays light
AppMarquee avoids keeping every full-resolution screen alive at once. The editor renders detail only around the active slide, uses schematic sidebar thumbnails, keeps a decode-only image cache, captures each export source once, and streams uncompressed PNGs into the ZIP. See Architecture.
Project storage
projects/
└── focus-garden/
├── project.json
└── assets/
├── 2b7d56c89bfa62e1.png
└── a8b701c50136ee9d.jpg
projects/ is ignored by git by default. Publishing this repository does not publish local screenshot projects.
Development
bun run check # unit tests + MCP bundle + production application build
bun run mcp:test # real stdio client smoke test (studio must be running)
bun run project:list
bun run appmarquee validate <project-id>
Useful references:
- Agent workflow
- Installation for Codex, Claude, and other clients
- Claude Desktop setup
- Architecture and security boundaries
- Contributing
Open-source lineage
AppMarquee is an independent open-source continuation of Parth Jadhav’s original app-store-screenshots project. The editor foundation remains MIT-licensed; AppMarquee adds a multi-project workspace, local project API, lower-memory rendering, revision-safe MCP tools, plugin packaging, and the new product identity.
License
MIT © AppMarquee contributors.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。