AppScreen

What the AppScreen MCP server does

The AppScreen MCP server lets an MCP-capable agent submit app captures and layout instructions to AppScreen. The service creates screenshot assets for App Store and Google Play listings, including supported device sizes and languages. The main input is a collection of screens, represented as layout JSON, together with image files.

A render may return a jobId for asynchronous processing or a download link for anonymous or inline use. Completed exports can be downloaded from a signed URL. These URLs expire after about one hour, so a later status poll can refresh access.

How it works

The local MCP process reads image data from paths on the machine where it runs. Pass each image as a file name and path, preferably using an absolute path or a path beginning with ~. Base64 is accepted for small synthetic test images, but the documented workflow is to pass real screenshots by path rather than embedding large encoded images in an agent request.

For asynchronous renders, call get_render_status with the returned job ID. The response reports the current state and includes a signed download URL when processing is complete. The service also returns quota metadata, including the current balance and, when available, the render cost.

Setup and configuration

Run the package without a separate installation using:

npx -y @appscreen/mcp

For Claude Desktop, add an appscreen MCP entry that runs npx -y @appscreen/mcp and optionally supplies APPSCREEN_API_KEY. Claude Code and Codex provide equivalent mcp add commands; Cursor uses the same JSON-style MCP configuration. The API key is an AppScreen MCP key beginning with mcm_live_, not the separate ask_live_ programmatic key.

APPSCREEN_API_KEY is optional. Without it, anonymous use is limited to five renders over the lifetime of the anonymous access. APPSCREEN_BASE_URL can override the default API endpoint, https://api.appscreen.co, and APPSCREEN_MCP_CLIENT can override the client identifier detected during MCP initialization.

Tools and capabilities

• render_screenshots submits layout JSON and image files. Authenticated calls return a job ID for keyed asynchronous work; anonymous or inline responses may include a download link directly.
• get_render_status checks a submitted job and returns its status and download URL when ready.
• Output can account for multiple screens, sizes, and languages. Export cost is based on the number of images produced: screens multiplied by sizes and languages.
• Signed-in workspaces use credits, where one exported image costs one credit. Failed exports are refunded, and automatic translation does not consume credits.
• Each response includes quota information. An export that exceeds the workspace balance fails with HTTP 402 and reports the cost and remaining balance.

Limitations and notes

Free signed-in usage provides 50 credits per month, one size per export, English, and up to 10 screens per export, with three renders per hour. Other plans increase the available credits, sizes, languages, screen count, or request rate as described by AppScreen. Anonymous access is capped at five lifetime renders rather than being metered through credits.

An MCP key spends credits from the workspace where that key was created. One-time credit packs are also available and remain valid for 12 months. When quota is low, responses include a notice; when it falls below 20% or reaches zero, an upgrade URL is included in metadata. The package is licensed under MIT.