lemgen-ai-design-mcp

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.

Related MCP server: Vision MCP Server

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

Example Workflows

1. Find A Visual Direction

User:

Find product photography inspiration for a matte black perfume bottle.

Agent:

1. Calls search_gallery(query="matte black perfume product photography").

2. Shows preview URLs and prompt IDs.

3. Calls get_inspiration(id=...) when the user picks one.

4. Adapts the prompt for the user's product.

2. Improve A Weak Prompt

User:

Make a poster for a new AI design tool.

Agent:

1. Calls enhance_prompt(idea="poster for a new AI design tool", style="poster").

2. Explains the visual direction.

3. Asks for approval before generation.

3. Generate An Image

User:

Looks good, generate it in 16:9.

Agent:

1. Calls generate_image(prompt=..., aspectRatio="16:9").

2. Returns the generation ID, status, and image URLs.

3. 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:

Model coverage:

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

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