Mock Screenshots MCP server

An MCP server that lets an AI agent compose a fake chat and get a deep link to the matching Mock Screenshots generator — pre-filled and ready to preview and download.

Output is watermarked and clearly fictional, intended for parody, education, design mockups and fiction. It is not for deception — see the ethics policy.

Tools

generate_fake_chat input

{
  "platform": "whatsapp",            // imessage | whatsapp | whatsapp-group | instagram | telegram | messenger | snapchat
  "contact": "Mom",                  // header name / username / group name
  "status": "typing…",               // header status line (optional)
  "device": "galaxy-s24",            // see list_devices (default iphone-16-pro)
  "dark": true,                      // dark mode (optional)
  "format": "image",                 // "image" (default) | "link" — see Screenshots section
  "messages": [
    {
      "text": "you home?",
      "sender": "them",
      "time": "19:01",
      "image": {
        "data": "<base64 PNG, JPEG, or WebP bytes>",
        "mimeType": "image/png",
        "alt": "A synthetic design mockup"
      }
    },
    { "text": "5 mins!", "sender": "me", "time": "19:02", "ticks": "read" }
  ]
}

messages[].image is optional and supported only for whatsapp and whatsapp-group, so existing text-only requests remain unchanged. Image-only messages are also supported. Attachments must be self-contained PNG, JPEG, or WebP bytes with a matching file signature; SVG, remote URLs, malformed base64, and files over 2 MB are rejected. Alternative text is limited to 160 characters, trimmed, and defaults to Attached image. The server also rejects more than 100 messages, message text over 4,000 characters, header metadata over 256 characters, and aggregate text/image source data over 6,000 bytes before it decodes attachments or builds duplicate JSON/base64url state. The merged site endpoint limits the complete base64url state to 8,000 characters, so the final encoded-state check remains authoritative and a smaller attachment may still be required after conversation metadata is included.

Attachment privacy and URL handling

Use only non-sensitive synthetic or already-public attachments. Attachment bytes are stored as a data URL inside the base64url-encoded JSON state in both the hosted GET /api/render?...&s=<state> URL and the generator edit ?s=<state> URL. Base64url is encoding, not encryption: anyone or any system with either URL can recover the attachment.

Treat every attachment-bearing render or edit URL as sensitive. It may be retained in MCP transcripts, client logs, browser history, proxy/CDN request logs, analytics/referrers, and cache keys. Do not use credentials, private user data, confidential screenshots, or private uploads as fixtures. The MCP's own preview request uses cache: "no-store" and referrerPolicy: "no-referrer" as defense in depth, but those client-side request settings cannot guarantee that the deployed endpoint, intermediaries, or a user-opened URL will not log or cache the full GET URL. No attachment or URL logging is added by this MCP server.

This residual exposure is inherent in the currently deployed URL-state contract. Roll back attachment support by reverting the attachment commit, or avoid messages[].image; existing text-only requests remain compatible and unchanged. A future opaque, expiring server-side state token would require a separate site/API contract and is intentionally not invented here.

Screenshots

Returns a rendered, watermarked screenshot server-side via the site's GET /api/render endpoint (Cloudflare Browser Rendering). Default output (format: "image") includes:

• Inline preview image (scaled, watermarked — displays immediately)
• Hosted full-res PNG URL (download / share in DMs, always watermarked and clearly fictional)
• Deep edit link to the generator, pre-filled with your conversation

Alternately, use format: "link" for text-only output (just the URLs, no image preview).

Always watermarked: Screenshots include a prominent "FAKE" watermark and cannot be disabled. This ensures they remain clearly fictional and non-deceptive for parody, education, design mockups and fiction — see the ethics policy.

Also returns a URL like https://mockscreenshots.com/fake-whatsapp-chat-generator?s=<state> that opens the generator with the conversation loaded (for preview/tweaking before final export).

Run

npm install
npm start           # stdio server

Use with Claude Desktop / Claude Code

// claude_desktop_config.json  (or: claude mcp add)
{
  "mcpServers": {
    "mockscreenshots": {
      "command": "node",
      "args": ["/absolute/path/to/mcp/server.mjs"]
    }
  }
}

Once published to npm you can instead use npx -y mockscreenshots-mcp.

Publishing / distribution

1. npm publish the mockscreenshots-mcp package.
2. Register on registry.modelcontextprotocol.io using server.json.
3. Submit to mcp.so, Smithery, PulseMCP, Glama, and open a PR to awesome-mcp-servers.

Each listing is a genuine dofollow dev-domain link + agent discovery — the uncontested distribution channel

How it works

The MCP server itself stays tiny and stateless: it builds URLs and encodes the conversation into a compact, URL-safe ?s= parameter (also read by the generator on load, src/lib/share.ts), then fetches a preview from the site's /api/render endpoint, which does the actual (always-watermarked) server-side rendering via Cloudflare Browser Rendering. This keeps the server dependency-light and keeps the human in the loop to preview, tweak and export.

The server never fetches a caller-supplied attachment URL. Preview responses are accepted only when they are PNG data from the fixed Mock Screenshots endpoint. The response body is consumed with a 10 MB streaming cap and the reader is cancelled immediately when that cap is crossed, even when Content-Length is absent or inaccurate. Endpoint errors, invalid responses, and timeouts safely fall back to the hosted image/edit links while retaining the watermark, ethics warning, and (when an attachment is present) attachment privacy warning.