agentsclimarketplace

Stitch mcp generate variants

Skill gabelul/stitch-kit/skills/stitch-mcp-generate-variants

Generates design variants of existing Stitch screens using the native variant API. Explore alternative layouts, color schemes, fonts, or content with configurable creativity levels.From its SKILL.md

Install
npx -y skills add gabelul/stitch-kit --skill stitch-mcp-generate-variants

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

SKILL.md

4.6 KB, ~1.1k tokens by cl100k_base, as published. Nobody here has run it

Stitch MCP — Generate Variants

Generates alternative versions of existing screens using Stitch's native variant generation API. This is more efficient than the text-prompt approach (1 API call vs. 3) and offers fine-grained control over what aspects to vary.

Critical prerequisite

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

You must have both a projectId AND at least one screenId of an existing screen. Variants are always based on an existing design — you can't generate variants from scratch.

When to use

  • User wants to explore alternative versions of an existing screen
  • After generation, user says "show me some variations" or "try different styles"
  • The orchestrator's Step 5b offers "Generate variants"
  • A/B testing — creating multiple options for stakeholder review

Call the MCP tool

{
  "name": "generate_variants",
  "arguments": {
    "projectId": "3780309359108792857",
    "selectedScreenIds": ["88805abc123def456"],
    "prompt": "Explore different color schemes while keeping the layout structure",
    "variantOptions": {
      "variantCount": 3,
      "creativeRange": "EXPLORE",
      "aspects": ["COLOR_SCHEME", "IMAGES"]
    },
    "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"]

The source screen(s) to generate variants from.

prompt — optional guidance for variant direction

Provide context about what kind of variations the user wants. The variantOptions do the heavy lifting, but the prompt adds nuance.

variantOptions — controls what and how much to vary

FieldTypeValuesDescription
variantCountint1–5Number of variants to generate
creativeRangeenumREFINE, EXPLORE, REIMAGINEHow much to deviate from the original
aspectsarraySee belowWhich design aspects to vary

creativeRange mapping from user language

User says→ creativeRangeWhat it does
"subtle changes", "minor tweaks", "polish"REFINESmall refinements, stays close to original
"alternatives", "different options", "explore"EXPLOREMeaningful differences while keeping the concept
"radical", "completely different", "reimagine"REIMAGINEMajor departures from the original design

aspects — what to vary

ValueVariesKeeps stable
LAYOUTStructure, spacing, component arrangementColors, fonts, content
COLOR_SCHEMEColors, gradients, contrastLayout, fonts, content
IMAGESPhotography, illustrations, iconsLayout, colors, text
TEXT_FONTTypography, font choices, sizesLayout, colors, content
TEXT_CONTENTCopy, labels, placeholder textLayout, colors, fonts

You can combine aspects: ["LAYOUT", "COLOR_SCHEME"] varies both simultaneously.

deviceType — optional

Same enum: MOBILE, DESKTOP, TABLET, AGNOSTIC

modelId — optional

ValueUse when
GEMINI_3_1_PRORecommended — complex layouts, high fidelity
GEMINI_3_FLASHFast iteration, wireframes, simple changes
GEMINI_3_PRODeprecated. Still works but will be removed. Use GEMINI_3_1_PRO instead.

Output

Returns new screens added to the project. Each variant appears as a separate screen in list_screens.

After generating variants

  1. Call stitch-mcp-list-screens to find all new variant screens
  2. Call stitch-mcp-get-screen for each to get screenshots
  3. Present side by side: "Here are your 3 variants: [screenshots]. Which one do you prefer?"
  4. Once the user picks a winner, offer:
    • "Edit the chosen variant further?" → stitch-mcp-edit-screens
    • "Convert to code?" → framework conversion workflow
    • "Generate more variants from the winner?" → another generate_variants call

Anti-patterns

  • Never generate variants without an existing screen — you need a source design
  • Never use projects/ID format for projectId or screenId — both must be numeric
  • Never set variantCount above 5 — the API caps at 5

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Gives 0 of the 12 instructions most mcp tooling skills give in ~1.1k tokens

Counted across 780 of the 1,136 authors here whose files we hold, read 2026-09-06

  • Use Zod for input validationin 34 of 780, across 21 files
  • Use stdio for local clientsin 27 of 780, across 10 files
  • Restart Claude Code after configurationin 26 of 780, across 23 files
  • Verify MCP server connection before using toolsin 23 of 780, across 17 files
  • Define input schemas for every toolin 20 of 780, across 11 files
  • Use Streamable HTTP for remote clientsin 18 of 780, across 8 files
  • Pin SDK version in package.jsonin 17 of 780, across 6 files
  • Keep server logic independent of transportin 16 of 780, across 6 files
  • Verify SDK methods against official documentationin 15 of 780, across 5 files
  • Format evaluation results as an XML filein 15 of 780, across 12 files
  • Test servers using the MCP Inspectorin 15 of 780, across 14 files
  • Create ten complex and independent evaluation questionsin 14 of 780, across 11 files

Said here and by no other author read

  • Use only when user mentions Stitch
  • Require a project ID and at least one screen ID
  • Call generate_variants tool with specified parameters
  • List screens after generating variants
  • Get screen details for each new variant
  • Present variants side by side to the user

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.

Keep looking

Skills are one crate of 325,949. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.