Agent Skill Integration Guide

This guide is written for AI agents, skill authors, and automation builders. The main goal is not to show a human how to call a REST endpoint manually. The goal is to help an agent add Mew.Design as a reusable design capability.

Use /open/api/design/generate when you are designing a Skill, tool, plugin, or workflow that should create visual designs on behalf of a user.

What the Skill Should Do

A Mew.Design Skill should let an agent turn a user’s design request into a structured design-generation call. It should:

• Recognize when the user wants a visual design, such as a poster, flyer, infographic, card, social post, business card, or menu.
• Ask for missing required assets, such as logos, QR codes, product photos, screenshots, or reference images.
• Convert the user’s request into a complete userQuery with exact text, purpose, audience, layout, and style requirements.
• Call Mew.Design through the Open API.
• Return the generated designUrl, explain what was created, and mention any failure clearly.
• Protect the API key and never expose it in final answers, logs, screenshots, or generated files.

Recommended Integration Path

Situation	Use
You are writing an agent Skill or tool wrapper	/open/api/design/generate
Your client already supports remote MCP servers	https://api.mew.design/mcp
A human wants to create manually	Mew.Design workspace

Default for Skills

For a reusable Skill, prefer the Open API. It gives the Skill a stable HTTP contract and lets the Skill decide when to ask questions, validate assets, call the endpoint, and present results.

Skill Installation Brief for Agents

If your environment supports installable skills, create a Skill named mew-design with a trigger description like this:

name: mew-design
description: Use when the user asks to create, redesign, generate, or refine visual designs such as posters, flyers, infographics, social posts, cards, menus, business cards, banners, invitations, or marketing graphics with Mew.Design.

The Skill body should include this workflow:

1. Collect the design brief: format, audience, purpose, exact copy, layout, style, size, and required assets.
2. If the user references a logo, QR code, product image, screenshot, or style reference but does not provide an accessible image URL, ask for it before calling the API.
3. Build userQuery as a complete design brief. Preserve exact user-provided text.
4. Use assetImages for elements that must appear in the final design.
5. Use referenceImage for style inspiration only.
6. Call POST https://api.mew.design/open/api/design/generate.
7. Return the generated design URL and a short summary.
8. If the call fails, explain whether the issue is auth, credits, invalid assets, timeout, or request validation.

Skill Configuration

The Skill needs a Mew.Design API key.

1. Ask the user to create an API key in Account Settings.
2. Store it as a secret, for example MEW_DESIGN_API_KEY.
3. Read it from the runtime environment at call time.
4. Never hard-code the key inside the Skill instructions or source code.

Secret handling

The API key consumes the user’s credits. A Skill must treat it like a password. Never print it, commit it, or include it in generated artifacts.

Open API Contract for the Skill

• Endpoint: POST https://api.mew.design/open/api/design/generate
• Auth: X-Api-Key: $MEW_DESIGN_API_KEY or Authorization: Bearer $MEW_DESIGN_API_KEY
• Content-Type: application/json
• Timeout guidance: allow up to 120 seconds for complex designs

Request Parameters

Parameter	Type	How the Skill should use it
userQuery	string	Required. Full design brief with exact copy, content hierarchy, purpose, audience, layout requirements, and constraints.
designConcept	string	Optional. Condensed visual direction: colors, typography, style, mood, layout approach.
referenceImage	object	Optional. Style reference only. Include a direct image url and optional tag.
width	number	Optional. 1 to 5000 pixels. Default 1080.
height	number	Optional. Use -1 for auto-adaptive height, or 100 to 5000 for fixed height.
assetImages	array	Optional. Images that must be integrated into the final design, such as logos, QR codes, product photos, or screenshots.

No fake URLs

If the user has not provided an image URL, do not invent placeholder URLs. Ask for the asset or continue without it only if the design can still be completed.

Example Skill Call

curl -X POST https://api.mew.design/open/api/design/generate \
  -H "X-Api-Key: $MEW_DESIGN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "userQuery": "Create an Instagram post for Maple Street Cafe. Include this exact text: \"Fall in Love with Every Sip\", \"20% Off All Week\", and \"Available at Maple Street Cafe\". Use warm autumn colors, a premium cafe style, and leave space for the logo in the top left.",
    "designConcept": "Warm autumn cafe campaign, premium typography, cozy product promotion, clean social media layout",
    "width": 1080,
    "height": 1080
  }'

Successful Response

{
  "flag": true,
  "statusCode": "B000000",
  "message": "success",
  "data": {
    "success": true,
    "designUrl": "https://...",
    "message": "Visual design generated successfully."
  }
}

The Skill should return data.designUrl to the user and summarize the generated design.

Agent Prompting Rules

When using the Skill, the agent should write requests like a designer brief, not like a one-line image prompt.

Good userQuery includes:

• Output format and channel
• Target audience and purpose
• Exact text to include
• Visual hierarchy and layout priorities
• Brand colors, mood, or typography style
• Assets that must appear
• Constraints, such as “do not change the logo” or “keep the QR code readable”

Bad userQuery:

Make a cool poster.

Better userQuery:

Create a vertical event poster for a student jazz night. Include the exact title "Midnight Brass", date "June 18", time "8 PM", venue "Blue Hall", and CTA "Reserve your seat". Use a moody navy and gold palette, large elegant typography, brass instrument accents, and a clean layout that keeps the date and venue readable from a distance.

Error Handling for Skills

The Skill should map failures into actionable messages:

• Missing or invalid key: Ask the user to configure MEW_DESIGN_API_KEY.
• Insufficient credits: Tell the user the request could not run because the account needs more credits.
• Invalid asset URL: Ask for a direct public image URL.
• Timeout: Suggest reducing complexity, fewer assets, or smaller fixed dimensions.
• Validation error: Show which field needs to be corrected.

Optional MCP Setup

If the agent platform already supports remote MCP servers, the user can configure Mew.Design directly:

• Server URL: https://api.mew.design/mcp
• Tool name: generateDesign
• Preferred auth header: X-Api-Key: your-api-key
• Compatible bearer auth: Authorization: Bearer your-api-key

{
  "mcpServers": {
    "mew-design": {
      "url": "https://api.mew.design/mcp",
      "headers": {
        "X-Api-Key": "your-api-key"
      }
    }
  }
}

MCP is not a plain REST endpoint

https://api.mew.design/mcp is for MCP-compatible clients. A Skill that wants normal JSON over HTTP should use /open/api/design/generate.

Credit Notes

Design generation follows the current Mew.Design credit rules:

• New design generation: 10 credits
• Design modification in the workspace: 5 credits
• Premium image models: 2-60 credits, depending on model and settings
• Download: no extra credits

Check pricing and credits before embedding cost assumptions in a public Skill description.

Ready to Install the Skill?

Create an API key, store it as MEW_DESIGN_API_KEY, then add the mew-design Skill workflow above to your agent environment.