MCP 服务器

lemgen-ai-design-mcp

Turns AI coding agents into visual creative assistants by providing a curated prompt gallery, prompt enhancement, and image/video generation via LemGen, with free inspiration tools and optional token-based generation.

README

<h1 align="center">LemGen AI Design MCP</h1>

Turn Claude Code, Cursor, Codex, Windsurf, Roo Code, and OpenClaw into a visual creative assistant. 4,533 curated AI image/video prompts · GPT Image · Nano Banana · Seedance · Midjourney · prompt enhancement · local preferences · LemGen generation tools

<a href="#quick-start">Quick Start</a> · <a href="#what-you-can-do">Tools</a> · <a href="#prompt-library">Prompt Library</a> · <a href="#client-setup">Client Setup</a> · <a href="README.zh-CN.md">中文</a>

Why This Exists

Most AI coding agents are good at writing text and code, but weak at visual creative work. LemGen MCP gives them a visual toolkit:

Search a real prompt gallery before inventing prompts from scratch.
Pull full prompts, images, model metadata, and source links into the agent context.
Expand rough ideas into production-grade prompts.
Generate images or videos through LemGen when the user approves the final prompt.
Keep all exploratory tools free and local-friendly; only generation needs a token.

The result is a cleaner workflow: browse inspiration, craft the direction, then generate without leaving your agent.

See The Library

Quick Start

Free inspiration tools work without any API key:

{
  "mcpServers": {
    "lemgen": {
      "command": "npx",
      "args": ["-y", "lemgen@latest"]
    }
  }
}

To enable generate_image and generate_video, create an API key at https://lemgen.org and add it to your MCP config:

{
  "mcpServers": {
    "lemgen": {
      "command": "npx",
      "args": ["-y", "lemgen@latest"],
      "env": {
        "LEMGEN_API_TOKEN": "sk_your_key_here"
      }
    }
  }
}

During local development, run from this repository:

pnpm install
pnpm build
node bin/lemgen-mcp.js

What You Can Do

Tool	Free	What it does	When to use it
`search_gallery`	Yes	Searches 4,533 curated prompts with preview media and source links.	The user wants ideas, examples, style references, or prompt recipes.
`get_inspiration`	Yes	Returns the full prompt, images/video, model, tags, stats, and LemGen URL.	The user picked an item from search results.
`enhance_prompt`	Yes	Turns a rough idea into a detailed image/video prompt.	The user says something short like "luxury perfume ad".
`list_models`	Yes	Lists supported image/video models, ratios, resolutions, and reference limits.	The user asks which model to use.
`prompt_tools`	Yes	Translates, polishes, or improves prompts.	The user already has a prompt but wants it cleaner.
`manage_preferences`	Yes	Stores local style, model, ratio, and favorite prompt preferences.	The user says "remember this style".
`generate_image`	Token	Submits an image generation job to LemGen.	The user approved a final image prompt.
`generate_video`	Token	Submits a video generation job to LemGen.	The user explicitly confirmed video generation.

Example Workflows

1. Find A Visual Direction

User:

Find product photography inspiration for a matte black perfume bottle.

Agent:

Calls search_gallery(query="matte black perfume product photography").
Shows preview URLs and prompt IDs.
Calls get_inspiration(id=...) when the user picks one.
Adapts the prompt for the user's product.

2. Improve A Weak Prompt

User:

Make a poster for a new AI design tool.

Agent:

Calls enhance_prompt(idea="poster for a new AI design tool", style="poster").
Explains the visual direction.
Asks for approval before generation.

3. Generate An Image

User:

Looks good, generate it in 16:9.

Agent:

Calls generate_image(prompt=..., aspectRatio="16:9").
Returns the generation ID, status, and image URLs.
Does not invent a visual description it cannot verify.

4. Generate A Video

Video is slower and more expensive, so the agent should confirm first:

This will create a LemGen video job. Continue?

After confirmation it calls generate_video.

Prompt Library

This repository ships with data/trending-prompts.json.

Current snapshot:

Type	Count
Image prompts	2,593
Video prompts	1,940
Total	4,533

Model coverage:

Model	Count
GPT Image	1,935
Seedance 2.0	1,940
Nano Banana Pro	489
Midjourney	104
Other image models	65

Each item includes:

type GalleryPrompt = {
  id: string;
  rank: number;
  title: string;
  prompt: string;
  author: string;
  model: string;
  categories: string[];
  image: string;
  images: string[];
  video_url?: string;
  media_type: "image" | "video";
  likes: number;
  views: number;
  source_url: string;
};

The companion data-only repository is:

https://github.com/aithink001/lemgen-trending-prompts

Client Setup

Claude Code Plugin

/plugin marketplace add aithink001/lemgen-ai-design-mcp
/plugin install lemgen@lemgen-marketplace

Restart Claude Code after installation.

Cursor

Add this to your Cursor MCP config:

{
  "mcpServers": {
    "lemgen": {
      "command": "npx",
      "args": ["-y", "lemgen@latest"],
      "env": {
        "LEMGEN_API_TOKEN": "sk_your_key_here"
      }
    }
  }
}

Codex

codex mcp add lemgen -- npx -y lemgen@latest

For generation, add LEMGEN_API_TOKEN to the MCP server environment in your Codex config.

Windsurf / Roo Code / Cline

Use the same MCP server block:

{
  "command": "npx",
  "args": ["-y", "lemgen@latest"],
  "env": {
    "LEMGEN_API_TOKEN": "sk_your_key_here"
  }
}

OpenClaw

Use the included skill:

openclaw/SKILL.md

It defines the recommended workflow for inspiration search, prompt enhancement, generation confirmation, and video safety.

Environment Variables

Variable	Default	Description
`LEMGEN_API_BASE_URL`	`https://lemgen.org`	LemGen API host.
`LEMGEN_API_TOKEN`	none	Required for image/video generation.
`LEMGEN_OUTPUT_DIR`	`~/Pictures/lemgen`	Reserved for future CLI image saving.
`LEMGEN_VIDEO_OUTPUT_DIR`	`~/Movies/lemgen`	Reserved for future CLI video saving.
`LEMGEN_PREFERENCES_PATH`	`~/.lemgen/preferences.json`	Local preferences file.

Development

pnpm install
pnpm validate
pnpm inspect

Validation runs:

TypeScript typecheck
Build
Brand scan
MCP client smoke test

Design Principles

Free discovery first; generation second.
Never generate expensive video without explicit confirmation.
Return actual URLs and IDs; do not hallucinate what an image looks like.
Keep prompt data source-linked to LemGen.
Keep public docs and data clean of legacy or duplicate-brand signals.

Roadmap

Publish the lemgen npm package.
Add a remote MCP endpoint at https://mcp.lemgen.org/mcp.
Add richer semantic search over the prompt library.
Add generated prompt packs for product photography, logos, portraits, posters, and video ads.
Add one-shot CLI generation commands for shell and CI workflows.

License

MIT

lemgen-ai-design-mcp

README

Why This Exists

See The Library

Quick Start

What You Can Do

Example Workflows

1. Find A Visual Direction

2. Improve A Weak Prompt

3. Generate An Image

4. Generate A Video

Prompt Library

Client Setup

Claude Code Plugin

Cursor

Codex

Windsurf / Roo Code / Cline

OpenClaw

Environment Variables

Development

Design Principles

Roadmap

License

推荐服务器