meigen-ai-design-mcp

Skill by ara.so — MCP Skills collection.

MeiGen AI Design MCP is an open-source MCP server that transforms AI coding tools (Claude Code, Cursor, Windsurf, Roo Code, OpenClaw, etc.) into professional design assistants. It provides unified access to 9 leading image and video generation models, a curated library of 1,446 trending prompts, and parallel sub-agent orchestration for batch generation.

What It Does

• Multi-provider image generation: MeiGen Cloud (9 models), OpenAI-compatible APIs, or local ComfyUI
• Video generation: Seedance 2.0, Happyhorse 1.0, Veo 3.1 via MeiGen platform
• Curated prompt library: 1,446 trending prompts from nanobanana-trending-prompts
• Smart prompt enhancement: Transform brief ideas into professional image prompts
• Parallel batch generation: Sub-agents for efficient multi-variant creation
• ComfyUI workflow management: Import, modify, and run custom workflows
• Standalone CLI mode: Use without an MCP host for scripts and CI/CD

Installation

Claude Code (Recommended)

# Via MeiGen marketplace
/plugin marketplace add jau123/MeiGen-AI-Design-MCP
/plugin install meigen@meigen-marketplace

# Or via wshobson/agents (30k+ stars)
/plugin marketplace add wshobson/agents
/plugin install meigen-ai-design@claude-code-workflows

Restart Claude Code after installation.

Other MCP-Compatible Hosts

Add to your MCP config (.mcp.json, claude_desktop_config.json, etc.):

{
  "mcpServers": {
    "meigen": {
      "command": "npx",
      "args": ["-y", "meigen@1.3.1"],
      "env": {
        "MEIGEN_API_TOKEN": "${MEIGEN_API_TOKEN}"
      }
    }
  }
}

Cursor / VS Code / Windsurf / Roo Code

npx meigen init cursor      # Cursor
npx meigen init vscode      # VS Code
npx meigen init windsurf    # Windsurf
npx meigen init roo         # Roo Code

OpenClaw

# Full plugin (commands + skills + MCP)
openclaw bundles install clawhub:meigen-ai-design

# Or skill only
npx clawhub@latest install creative-toolkit

Hermes Agent

Add to ~/.hermes/config.yaml:

mcp_servers:
  meigen:
    command: "npx"
    args: ["-y", "meigen@1.3.1"]
    env:
      MEIGEN_API_TOKEN: "${MEIGEN_API_TOKEN}"
    timeout: 600          # Video generation can take 5-10 min
    connect_timeout: 120  # First npx download needs time

Configuration

Initial Setup

Run the setup wizard:

/meigen:setup

The wizard walks through:

1. Provider selection: ComfyUI (local), MeiGen Cloud, or OpenAI-compatible API
2. Credentials: API tokens, URLs, or keys
3. Restart: One final restart to activate

Provider Options

Option 1: Local ComfyUI (Free)

Run on your own GPU with full control:

{
  "comfyuiUrl": "http://localhost:8188",
  "comfyuiDefaultWorkflow": "txt2img"
}

Environment variables:

COMFYUI_URL=http://localhost:8188
COMFYUI_DEFAULT_WORKFLOW=txt2img

Option 2: MeiGen Cloud

Get your API token at meigen.ai → Settings → API Keys:

{
  "meigenApiToken": "meigen_sk_..."
}

Environment variable:

MEIGEN_API_TOKEN=meigen_sk_...

Option 3: OpenAI-Compatible API

Bring your own key and endpoint:

{
  "openaiCompatibleApiKey": "sk-...",
  "openaiCompatibleEndpoint": "https://api.openai.com/v1"
}

Environment variables:

OPENAI_API_KEY=sk-...
OPENAI_API_BASE=https://api.openai.com/v1

MCP Tools

search_gallery (Free)

Search 1,446 curated trending prompts with visual previews.

TypeScript usage:

const result = await use_mcp_tool({
  server_name: "meigen",
  tool_name: "search_gallery",
  arguments: {
    keywords: "product photography luxury",
    limit: 10
  }
});

Natural language prompt:

"Search the gallery for luxury product photography inspiration"

get_inspiration (Free)

Get full prompt, all images, and metadata for a specific gallery entry.

TypeScript usage:

const inspiration = await use_mcp_tool({
  server_name: "meigen",
  tool_name: "get_inspiration",
  arguments: {
    id: "nanobanana_12345"
  }
});

enhance_prompt (Free)

Transform brief ideas into professional image prompts with style awareness.

TypeScript usage:

const enhanced = await use_mcp_tool({
  server_name: "meigen",
  tool_name: "enhance_prompt",
  arguments: {
    userPrompt: "cat in kitchen",
    style: "photorealistic",
    additionalContext: "sunlit, warm tones"
  }
});

Natural language prompt:

"Enhance this prompt: a cat in a kitchen"

list_models (Free)

List available models across all configured providers.

TypeScript usage:

const models = await use_mcp_tool({
  server_name: "meigen",
  tool_name: "list_models",
  arguments: {}
});

generate_image (Requires API Key)

Generate an image with automatic provider routing, reference image upload, and aspect ratio handling.

TypeScript usage:

const image = await use_mcp_tool({
  server_name: "meigen",
  tool_name: "generate_image",
  arguments: {
    prompt: "a minimalist tech logo, geometric shapes, blue gradient",
    model: "midjourney-v8.1",
    aspectRatio: "1:1",
    referenceImagePath: "/Users/dev/Desktop/sketch.jpg"  // Optional
  }
});

Natural language prompt:

"Generate a minimalist tech logo in 1:1 ratio using Midjourney V8.1"

With reference image:

"Generate a product shot based on this bottle.jpg in my Downloads folder"

generate_video (Requires API Key)

Generate video via MeiGen platform (Seedance 2.0, Happyhorse 1.0, or Veo 3.1).

TypeScript usage:

const video = await use_mcp_tool({
  server_name: "meigen",
  tool_name: "generate_video",
  arguments: {
    prompt: "Camera slowly zooms into the perfume bottle, bokeh background",
    model: "seedance-2.0-fast",
    firstFrameImagePath: "/Users/dev/Pictures/bottle.jpg",  // Optional
    duration: 5
  }
});

Natural language prompt:

"Animate this product photo into a 5-second video with a slow zoom"

comfyui_workflow (Free)

Manage ComfyUI workflow templates.

TypeScript usage:

// List workflows
const workflows = await use_mcp_tool({
  server_name: "meigen",
  tool_name: "comfyui_workflow",
  arguments: {
    action: "list"
  }
});

// View a workflow
const workflow = await use_mcp_tool({
  server_name: "meigen",
  tool_name: "comfyui_workflow",
  arguments: {
    action: "view",
    name: "txt2img"
  }
});

// Import a workflow
await use_mcp_tool({
  server_name: "meigen",
  tool_name: "comfyui_workflow",
  arguments: {
    action: "import",
    name: "custom-flux",
    workflow: JSON.stringify(workflowJson)
  }
});

manage_preferences (Free)

Store user preferences (style, aspect ratio, model, favorite prompts).

TypeScript usage:

await use_mcp_tool({
  server_name: "meigen",
  tool_name: "manage_preferences",
  arguments: {
    action: "set",
    key: "preferredStyle",
    value: "minimalist editorial"
  }
});

const prefs = await use_mcp_tool({
  server_name: "meigen",
  tool_name: "manage_preferences",
  arguments: {
    action: "get"
  }
});

Slash Commands

Quick shortcuts for common tasks:

Example:

/meigen:gen a calico cat in a sunlit kitchen, photorealistic
/meigen:find luxury product photography
/meigen:models

Standalone CLI Mode

Use MeiGen without an MCP host (for shell scripts, CI/CD, etc.):

# Set token once
export MEIGEN_API_TOKEN=meigen_sk_...

# Basic generation
npx meigen gen --prompt "a calico cat in a sunlit kitchen"

# Specific model + aspect ratio
npx meigen gen -p "tech logo" -m midjourney-v8.1 -r 1:1

# With reference image
npx meigen gen -p "product hero shot" --ref ~/Desktop/bottle.jpg

# Submit only (no polling, good for CI)
npx meigen gen -p "..." --no-wait

# Machine-readable output (pipe to jq)
npx meigen gen -p "..." --json | jq -r '.imageUrls[0]'

Full CLI flags:

npx meigen gen --help

Options:
  -p, --prompt <text>        Image prompt (required)
  -m, --model <name>         Model name (default: nanobanana-2)
  -r, --ratio <ratio>        Aspect ratio (e.g. 16:9, 1:1, 3:4)
  --ref <path>               Reference image path
  --no-wait                  Submit only, don't poll for result
  --json                     Machine-readable JSON output
  -h, --help                 Display help

Output:

• Images saved to ~/Pictures/meigen/ (override with MEIGEN_OUTPUT_DIR)
• Videos saved to ~/Movies/meigen/

Common Patterns

Parallel Batch Generation

Generate multiple variations in parallel using sub-agents:

Natural language prompt:

"Create 4 product display images for this perfume, one of which should feature a model"

The AI will:

1. Upload the reference image
2. Craft 4 distinct prompts (luxury still life, model campaign, botanical, minimalist)
3. Spawn 4 image-generator sub-agents in parallel
4. Return all 4 images in ~2 minutes

Style-Aware Prompt Enhancement

Natural language prompt:

"Enhance this for a cinematic feel: a coffee cup on a desk"

Enhanced result:

A pristine white ceramic coffee cup on a dark walnut desk, steam rising 
gracefully, shot with shallow depth of field, warm morning light streaming 
through venetian blinds creating dramatic shadows, cinematic color grading, 
shot on Arri Alexa, 35mm lens, f/1.4

ComfyUI Workflow Import

Import any ComfyUI API-format workflow:

// Read workflow from file
const workflowJson = JSON.parse(
  fs.readFileSync('path/to/workflow_api.json', 'utf8')
);

// Import into MeiGen
await use_mcp_tool({
  server_name: "meigen",
  tool_name: "comfyui_workflow",
  arguments: {
    action: "import",
    name: "flux-lora-custom",
    workflow: JSON.stringify(workflowJson)
  }
});

// Use it
await use_mcp_tool({
  server_name: "meigen",
  tool_name: "generate_image",
  arguments: {
    prompt: "cyberpunk street scene",
    model: "comfyui",
    workflow: "flux-lora-custom"
  }
});

Video from Still Image

Animate a product photo into a 5-second video:

const video = await use_mcp_tool({
  server_name: "meigen",
  tool_name: "generate_video",
  arguments: {
    prompt: "Slow camera push towards product, bokeh background, cinematic lighting",
    model: "seedance-2.0-fast",
    firstFrameImagePath: "/Users/dev/Desktop/product.jpg",
    duration: 5
  }
});

Models:

• seedance-2.0-fast: 5s, 1280×720, ~30s generation
• seedance-2.0-pro: 5s, 1280×720, higher quality, ~2min
• happyhorse-1.0: 6s, 1360×768, realistic motion
• veo-3.1: 8s+, 1280×720, ultra-realistic, ~5-10min

Multi-Model Comparison

Generate the same prompt with different models:

Natural language prompt:

"Generate this prompt with Nanobanana 2, Midjourney V8.1, and Flux 2 Klein: a serene Japanese garden at dawn"

The AI will parallelize across 3 sub-agents and present all results.

Troubleshooting

"No provider configured"

Symptom: generate_image fails with "No provider configured"

Solution:

/meigen:setup
# Or set environment variables manually:
export MEIGEN_API_TOKEN=meigen_sk_...

ComfyUI connection refused

Symptom: ECONNREFUSED 127.0.0.1:8188

Solution:

1. Ensure ComfyUI is running: http://localhost:8188
2. Check firewall isn't blocking the port
3. Verify COMFYUI_URL matches your actual URL

Reference image upload fails

Symptom: Local image path not found

Solution:

• Use absolute paths: /Users/username/Desktop/image.jpg
• Or relative to project root: ./assets/image.jpg
• MeiGen auto-compresses images >5MB before upload

Video generation timeout

Symptom: MCP tool times out before video finishes

Solution:

• For Hermes Agent, set timeout: 600 in config
• For other hosts, increase MCP server timeout (default 120s → 600s)
• Veo 3.1 can take 5-10 minutes for high-quality 8s videos

"Model not available"

Symptom: Requested model not found

Solution:

/meigen:models
# Or use tool:
use_mcp_tool({
  server_name: "meigen",
  tool_name: "list_models",
  arguments: {}
})

Available models depend on your provider:

• MeiGen Cloud: nanobanana-2, gpt-image-2, seedream-5.0, midjourney-v8.1, flux-2-klein, seedance-2.0-fast, seedance-2.0-pro, happyhorse-1.0, veo-3.1
• ComfyUI: Uses local workflows, model specified in workflow JSON
• OpenAI-compatible: dall-e-3, dall-e-2, or custom models at your endpoint

Prompt library search returns no results

Symptom: search_gallery finds nothing

Solution:

• Try broader keywords: "product" instead of "luxury minimalist product photography"
• Use style terms: "cinematic", "minimalist", "vintage", "3D render"
• Browse without keywords (empty search) to see all 1,446 prompts

Environment Variables Reference

Resources

• GitHub: jau123/MeiGen-AI-Design-MCP
• Homepage: meigen.ai
• Prompt Library: nanobanana-trending-prompts
• Discord: Join community
• License: MIT