Skip to content
Development
Skill

/stitch-mcp-edit-screens

Edits existing Stitch screens with text prompts — the iteration tool. Change colors, layout, content, or style without regenerating from scratch. Supports refinement loops via output_components suggestions.

From plugin
stitch-kit
4536 skills1 agent2 hooks
Install
$ npx -y skills add gabelul/stitch-kit --skill stitch-mcp-edit-screens --agent claude-code

How it fires

How this skill gets triggered: by you, by Claude, or both.

  • Fires itselfAuto-invocation. Claude auto-loads it when your prompt matches the work.Auto-invocation is when the right skill fires by itself at the right moment, driven by a FLOW.md router and a hook, instead of you invoking it by name. It is the difference between a skill being installed and a skill actually getting used.Read the full definition →
  • You can call itInvoke it directly when you want it.
  • Slash command/stitch-mcp-edit-screens

Context preview

The summary Claude sees to decide when to auto-load this skill.

Edits existing Stitch screens with text prompts — the iteration tool. Change colors, layout, content, or style without regenerating from scratch. Supports refinement loops via output_components suggestions.

SKILL.md

stitch-mcp-edit-screens.SKILL.md
name: stitch-mcp-edit-screens
description: Edits existing Stitch screens with text prompts — the iteration tool. Change colors, layout, content, or style without regenerating from scratch. Supports refinement loops via output_components suggestions.
allowed-tools:
  - "stitch*:*"
  - "Bash"
  - "Read"
  - "Write"

Stitch MCP — Edit Screens

Edits one or more existing screens using a text prompt. This is the primary iteration tool — instead of regenerating a screen from scratch (60–180s), you can refine what you already have with targeted changes.

Critical prerequisite

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

You must have both a `projectId` AND at least one `screenId` before calling this. If you don't:

  • List screens via `stitch-mcp-list-screens` to find screen IDs
  • Parse from context: `projects/123/screens/456` → projectId: `123`, screenId: `456`

When to use

  • User wants to modify an existing Stitch screen (change colors, layout, content, style)
  • After generation, user says "make it darker", "change the font", "move the nav to the side"
  • The orchestrator's Step 5b offers "Edit this screen"
  • Any iterative refinement on a Stitch design

Call the MCP tool

{
  "name": "edit_screens",
  "arguments": {
    "projectId": "3780309359108792857",
    "selectedScreenIds": ["88805abc123def456"],
    "prompt": "Change the background to dark mode (#09090B). Make the primary color indigo (#818CF8). Increase the font size of the header to 32px bold.",
    "deviceType": "DESKTOP",
    "modelId": "GEMINI_3_1_PRO"
  }
}

Parameter reference

`projectId` — numeric ID only, no prefix

✅ "3780309359108792857"
❌ "projects/3780309359108792857"

`selectedScreenIds` — array of numeric screen IDs

✅ ["88805abc123def456"]
❌ ["projects/123/screens/88805abc123def456"]
❌ ["screens/88805abc123def456"]

Select one or more screens to edit. All selected screens receive the same edit instruction.

`prompt` — the edit instruction

Apply the same quality bar as generation prompts:

  • Be specific: "Change background to #09090B" not "make it darker"
  • Name exact components: "the header navigation" not "the top part"
  • Include values: hex colors, px sizes, specific content text
  • One focused change per call produces better results than many changes at once

`deviceType` — optional

Same enum as `generate_screen_from_text`: `MOBILE`, `DESKTOP`, `TABLET`, `AGNOSTIC`

`modelId` — optional

| Value | Use when | |-------|---------| | `GEMINI_3_1_PRO` | **Recommended** — complex layouts, high fidelity | | `GEMINI_3_FLASH` | Fast iteration, wireframes, simple changes | | `GEMINI_3_PRO` | **Deprecated.** Still works but will be removed. Use `GEMINI_3_1_PRO` instead. |

Handling `output_components`

The response may contain `output_components` with suggestions:

{
  "outputComponents": [
    { "text": "I've updated the background color and adjusted contrast ratios." },
    { "suggestion": "Would you also like me to update the sidebar to match the dark theme?" }
  ]
}

**When you see suggestions:** 1. Present them to the user 2. If the user accepts: call `edit_screens` again with the suggestion as the new `prompt` 3. This creates a natural refinement loop — keep going until the user is satisfied

Timing

Same as generation: 60–180 seconds is normal.

  • Do NOT retry during this window
  • Do NOT assume failure if it takes > 60 seconds
  • Each call creates a new edit — retries mean duplicate edits

After editing

1. Re-fetch the screen: `stitch-mcp-get-screen` with the same projectId and screenId 2. Show the updated screenshot to the user 3. Offer:

  • "Continue editing?" → another `edit_screens` call
  • "Generate variants of this version?" → `stitch-mcp-generate-variants`
  • "Convert to code?" → framework conversion workflow

Anti-patterns

  • **Never send vague edit prompts** — "make it better" will produce unpredictable results
  • **Never use `projects/ID` format** for projectId or screenId — both must be numeric
  • **Never batch unrelated edits** — "change the color AND completely redo the layout" works poorly
Read more
Ships withstitch-kit

Your coding agent writes decent code and designs terrible UI. stitch-kit fixes the second half — it wires agents into Google Stitch (text prompts → genuinely beautiful screens) and teaches them to drive it properly.

Get the whole plugin

Other skills on stitch-kit.