Stitch MCP Get Screen

Stitch MCP — Get Screen

Retrieves the full output of a Stitch-generated screen: the HTML source code URL and the screenshot image URL. This is the gateway between Stitch design and framework conversion.

Critical prerequisite

Only use this skill when the user explicitly mentions "Stitch".

If you already have both projectId AND screenId, call this directly — do not call get_project first.

When to use

• After list_screens has returned a screenId
• User provides a Stitch screen URL and wants to convert it to code
• Retrieving assets for any conversion skill: Next.js, Svelte, HTML, React Native, or SwiftUI

Step 1: Parse IDs from context

The user may provide the screen reference in different formats:

Step 2: Call the MCP tool

get_screen is mid-migration: name is the current identifier and takes the full resource path, while projectId and screenId are marked deprecated but are still required. Send all three.

{
  "name": "get_screen",
  "arguments": {
    "name": "projects/3780309359108792857/screens/88805abc123def456",
    "projectId": "3780309359108792857",
    "screenId": "88805abc123def456"
  }
}

The prefix rule differs per field, which is the whole reason this wrapper exists:

✅ name:      "projects/3780309359108792857/screens/88805abc123def456"
❌ name:      "88805abc123def456"

✅ projectId: "3780309359108792857"
❌ projectId: "projects/3780309359108792857"

✅ screenId:  "88805abc123def456"
❌ screenId:  "screens/88805abc123def456"

name is the full path; the other two are bare. Omitting name fails — it's required despite the deprecated pair still being mandatory too.

Output schema

{
  "name": "projects/3780309359108792857/screens/88805abc123def456",
  "htmlCode": {
    "downloadUrl": "https://storage.googleapis.com/stitch-output/..."
  },
  "screenshot": {
    "downloadUrl": "https://storage.googleapis.com/stitch-screenshots/..."
  },
  "figmaExport": {
    "downloadUrl": "https://storage.googleapis.com/stitch-figma/..."
  },
  "width": 390,
  "height": 844,
  "deviceType": "MOBILE"
}

Step 3: Download the HTML reliably

AI fetch tools frequently fail on Google Cloud Storage URLs. Use the bash script:

bash scripts/fetch-stitch.sh "[htmlCode.downloadUrl]" "temp/source.html"

Always quote the URL to handle special characters.

Step 4: Determine framework and route to conversion

After retrieving the screen, check what the user wants to do with it.

Check the deviceType in the response first — it shapes which options to suggest:

Then route based on user intent:

References

• scripts/fetch-stitch.sh — Reliable GCS HTML downloader