MCP 服务器

AI Image-Gen MCP Server

Transforms DALL-E 3, DALL-E 2, and GPT-Image-1 into MCP-compatible tools for generating images from natural language descriptions, supporting batch creation, style control, and flexible sizing.

README

AI Image‑Gen MCP Server

Version 0.1.0 – MVP public release
Conforms to the Model Context Protocol spec (2025‑06‑18).

What's this?

A production‑ready MCP server that transforms state‑of‑the‑art image generators into plug‑and‑play tools for any MCP‑aware client. Currently shipping with DALL·E 3, DALL·E 2, and experimental GPT‑Image‑1 – all accessible through a unified interface.

Why MCP?

MCP is the USB‑C of AI context: one protocol, endless integrations. Ship one server, hook it into Claude Desktop, Claude Code, VS Code, or your own chatbot – the host handles UI, auth, and conversation flow.

Quick examples

You ask	The server delivers
"Design a cyberpunk logo for my startup"	High‑res PNG via DALL·E 3 with style presets
"Generate 5 variations of this product shot"	Batch generation via DALL·E 2 (n=5 support)
"Create concept art for a steampunk airship"	Artistic rendering with metadata and prompt history

If you can describe it, we can render it. 💫

Core MCP Concepts

This server implements all three MCP primitives:

Tools – generate_image with model selection, size, and style options
Resources – Available models and their capabilities exposed as MCP resources
Prompts – Built‑in templates for product_mockup and concept_art workflows

Feature Highlights

Multi‑Model Support – DALL·E 3 (default), DALL·E 2, and GPT‑Image‑1 via unified API
Smart Storage – Local cache with timestamped filenames and JSON metadata
Flexible Sizing – From 256×256 thumbnails to 1792×1024 widescreen masterpieces
Style Control – vivid or natural rendering (DALL·E 3)
Batch Generation – Create up to 10 variations per prompt (DALL·E 2)
Claude Integration – First‑class support for Desktop and Code editions

Architecture

graph TD
    Client["MCP Client (Claude Desktop/Code)"] -- JSON‑RPC 2.0 --> Server["Image‑Gen MCP Server"]
    Server --> Router["Model Router"]
    Router -->|OpenAI API| DALLE3["DALL·E 3"]
    Router -->|OpenAI API| DALLE2["DALL·E 2"]
    Router -->|Responses API| GPT["GPT‑Image‑1"]
    Server --> Storage["Local Storage + Metadata"]
    Storage --> Client

Quickstart

Prerequisites

Python 3.11+
OpenAI API key
Claude Desktop or Claude Code (for MCP integration)

Installation

git clone https://github.com/krystian-ai/ai-image-gen-mcp.git
cd ai-image-gen-mcp
python3.11 -m venv .venv && source .venv/bin/activate
pip install -e ".[image,dev]"

Configuration

cp .env.example .env
# Edit .env and add your OpenAI API key

Key settings:

OPENAI_API_KEY=sk-...
MODEL_DEFAULT=dall-e-3
CACHE_DIR=/tmp/ai-image-gen-cache

Run Standalone

# Via MCP CLI
mcp-imageserve stdio

# Direct execution
python -m ai_image_gen_mcp.server --transport=stdio

Claude Desktop Integration

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "ai-image-gen": {
      "command": "/path/to/ai-image-gen-mcp/.venv/bin/python",
      "args": ["-m", "ai_image_gen_mcp.server", "stdio"],
      "transport": "STDIO",
      "env": {
        "PYTHONPATH": "/path/to/ai-image-gen-mcp/src",
        "OPENAI_API_KEY": "your-api-key-here"
      }
    }
  }
}

Claude Code Integration

Option 1: Project‑specific `.mcp.json` (Recommended)

Drop this in your project root:

{
  "ai-image-gen": {
    "command": "python",
    "args": ["-m", "ai_image_gen_mcp.server", "stdio"],
    "transport": "STDIO",
    "env": {
      "PYTHONPATH": "src",
      "OPENAI_API_KEY": "your-api-key-here"
    }
  }
}

Claude Code auto‑detects and loads it. ✨

Option 2: Global Config

Add to ~/.config/claude-code/settings.json for system‑wide access.

Model Capabilities

Model	Sizes	Styles	Batch (n)	Speed	Notes
DALL·E 3	1024×1024, 1792×1024, 1024×1792	vivid, natural	1	Fast	Best quality, default model
DALL·E 2	256×256, 512×512, 1024×1024	N/A	1-10	Fast	Good for variations
GPT‑Image‑1	Fixed (model‑determined)	N/A	1	Slow (20s+)	Experimental, may timeout

Usage Examples

Basic Generation

Generate a minimalist logo for a productivity app

With Parameters

Create a vivid 1792x1024 banner of a futuristic cityscape using dall-e-3

Batch Creation

Generate 5 variations of a coffee cup product photo using dall-e-2

Interactive HTML Demo

Explore the server's capabilities through our interactive web interface:

cd examples/html
open index.html  # macOS
# or
xdg-open index.html  # Linux
# or just open in your browser

The demo showcases:

Live Examples – Generated images with their prompts
Model Comparison – See outputs from DALL·E 3, DALL·E 2, and GPT-Image-1
Interactive Gallery – Carousel of stunning AI-generated artwork
Integration Guide – How to connect with Claude Desktop/Code

Perfect for visualizing what's possible before diving into the API!

Development

Testing

pytest                           # Full suite
pytest --cov=ai_image_gen_mcp   # Coverage report
python test_dalle.py            # Live API test

Code Quality

black src/        # Format
ruff check src/   # Lint
mypy src/         # Type check

Project Structure

ai-image-gen-mcp/
├── src/ai_image_gen_mcp/
│   ├── server.py          # FastMCP server entry
│   ├── models/            # Model implementations
│   └── config.py          # Environment config
├── tests/                 # Comprehensive test suite
├── examples/
│   └── html/              # Interactive web demo
├── assets/                # Logo images
└── .mcp.json             # Claude Code config

Troubleshooting

Issue	Solution
MCP not detected	Ensure `.mcp.json` exists in project root
API key errors	Check `OPENAI_API_KEY` in `.env` or environment
Import errors	Verify `PYTHONPATH` includes `src/` directory
GPT‑Image‑1 timeouts	Known issue – use DALL·E models for reliability
Claude Desktop issues	Use full paths to venv Python executable

Roadmap

Version	Focus	Status
0.1	MVP with 3 models, local storage	✅ Shipped
0.2	S3/GCS storage, signed URLs	🚧 Planning
0.3	Stable Diffusion, ComfyUI integration	📋 Backlog
0.4	Inpainting, upscaling, style transfer	💭 Ideas