Creative Toolkit

Creative Toolkit

Generate professional AI images through a unified interface that routes across multiple providers. Search curated prompts, enhance ideas into production-ready descriptions, and manage local ComfyUI workflows — all from a single MCP server.

Quick Start

Add the MCP server to your mcporter config (~/.config/mcporter/config.json):

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

Set your API token in ~/.clawdbot/.env or shell environment:

export MEIGEN_API_TOKEN="meigen_sk_..."

Generate your first image:

mcporter call creative-toolkit.generate_image prompt="a minimalist perfume bottle on white marble, soft directional lighting, product photography"

Or try it without any config (ad-hoc stdio mode):

mcporter call --stdio "npx -y meigen@latest" generate_image prompt="a ceramic vase with morning light"

No API key? Free tools still work:

mcporter call creative-toolkit.search_gallery query="cyberpunk"
mcporter call creative-toolkit.enhance_prompt brief="a cat in space" style="realistic"

Setup

Get an API Token

1. Visit meigen.ai → sign in → click avatar → Settings → API Keys
2. Create a new key (starts with meigen_sk_)
3. Set as environment variable or save to config:

# Shell environment or ~/.clawdbot/.env
export MEIGEN_API_TOKEN="meigen_sk_..."

Or save to ~/.config/meigen/config.json:

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

Alternative Providers

You can use your own OpenAI-compatible API or a local ComfyUI instance instead of — or alongside — the default provider. Save to ~/.config/meigen/config.json:

OpenAI / Together AI / Fireworks AI:

{
  "openaiApiKey": "sk-...",
  "openaiBaseUrl": "https://api.together.xyz/v1",
  "openaiModel": "black-forest-labs/FLUX.1-schnell"
}

Local ComfyUI:

{
  "comfyuiUrl": "http://localhost:8188"
}

Import workflows with the comfyui_workflow tool (action: import). The server auto-detects key nodes (KSampler, CLIPTextEncode, EmptyLatentImage) and fills in prompt, seed, and dimensions at runtime.

Multiple providers can be configured simultaneously. Auto-detection priority: MeiGen > ComfyUI > OpenAI.

Available Tools

Free — no API key required

Requires configured provider

Usage Patterns

Basic generation

Generate a product photo of a ceramic vase with morning light

The server picks the best provider, generates the image, and returns a URL + local file path.

Prompt enhancement then generation

For brief ideas, enhance first for much better results:

1. enhance_prompt brief="futuristic city" style="realistic"
   → Returns detailed prompt with camera lens, lighting setup, atmospheric effects

2. generate_image prompt="<enhanced prompt>" aspectRatio="16:9"
   → Generates with the enhanced prompt

Reference image workflow

Use an existing image to guide style and composition:

1. upload_reference_image filePath="~/Desktop/my-logo.png"
   → Returns public URL

2. generate_image prompt="coffee mug mockup with this logo" referenceImages=["<url>"]
   → Generates using the reference for style guidance

Reference images work across all providers.

Gallery exploration

Semantic search understands intent — "dreamy portrait with soft light" finds relevant results even without exact keyword matches:

1. search_gallery query="dreamy portrait with soft light"
   → Finds semantically similar prompts with thumbnails

2. search_gallery category="Product"
   → Browse by category from 1,300+ curated prompts

3. get_inspiration id="<entry_id>"
   → Get full prompt text — copy and modify for your own generation

ComfyUI workflows

1. comfyui_workflow action="list"
   → See saved workflows

2. comfyui_workflow action="view" name="txt2img"
   → See adjustable parameters (steps, CFG, sampler, checkpoint)

3. comfyui_workflow action="modify" name="txt2img" modifications={"steps": 30, "cfg": 7.5}
   → Adjust without editing JSON

4. generate_image prompt="..." workflow="txt2img"
   → Generate using the custom workflow

Provider Comparison

Prompt Enhancement Styles

enhance_prompt supports three style modes, each producing different types of detail:

Security & Privacy

Remote package execution: This skill runs as an MCP server via npx meigen@latest. The package is published on npmjs.com under the meigen name with full source code available at GitHub. No code is obfuscated or minified beyond standard TypeScript compilation.

Reference image upload: The upload_reference_image tool compresses local images (max 2MB, 2048px) and uploads them to a CDN for use as generation references. This is always user-initiated — the tool never accesses or uploads files without explicit invocation. Uploaded images are used solely as style references for image generation.

API tokens: MEIGEN_API_TOKEN is stored locally in environment variables or ~/.config/meigen/config.json with chmod 600 permissions. Tokens are only sent to the configured provider's API endpoint and never logged or transmitted elsewhere.

No telemetry: The MCP server does not collect analytics, usage data, or send any information to third parties beyond the configured image generation provider.

Troubleshooting

"No image generation providers configured" → Set MEIGEN_API_TOKEN or configure an alternative provider in ~/.config/meigen/config.json

Timeout during generation → Image generation typically takes 10-30 seconds. During high demand, it may take longer. The server polls with a 5-minute timeout.

ComfyUI connection refused → Ensure ComfyUI is running and accessible at the configured URL. Test with: curl <url>/system_stats

"Model not found" → Run list_models to see available models for your configured providers.

Reference image rejected → Images must be public URLs (not local paths). Use upload_reference_image to convert local files to URLs first.