Image Gen MCP

⚡ Introduction

Image Gen MCP is a STDIO MCP server for generating images with OpenAI and Google Gemini. It lets AI coding tools create image assets and save them directly inside your project, without a separate image generation workflow or manual downloads.

✨ Features

• 🧠 Two AI providers - Generate images with OpenAI or Google Gemini

• 🖼️ Batch generation - Create several images in one request, with up to 10 jobs running at once

• 🎨 Image controls - Set size, quality, aspect ratio, format, and background when supported by the model

• 💾 Explicit file output - Save generated images to a required absolute directory

• 🏷️ Clear filenames - Files include the requested name, provider, model, and timestamp

• 📋 Useful results - Get a summary of saved files and any jobs that failed

⚙️ Tech Stack

• Language - TypeScript

• Runtime - Node.js 20 or newer

• MCP - Model Context Protocol server over STDIO

• AI SDKs - OpenAI and Google Gen AI

• Validation - Zod

• Build tools - tsup and Biome

📦 Installation

Requirements:

• Node.js 20 or newer

• An OpenAI API key, a Gemini API key, or both

• An MCP-compatible client

Add the following server configuration to your MCP client. The exact location of this configuration depends on the client.

{
  "mcpServers": {
    "image-gen-mcp": {
      "command": "npx",
      "args": ["-y", "@s4shibam/image-gen-mcp"],
      "env": {
        "OPENAI_API_KEY": "your-openai-api-key",
        "GEMINI_API_KEY": "your-gemini-api-key"
      }
    }
  }
}

You only need the API key for the provider you want to use. You can also place the keys in a .env file in the workspace where the MCP server runs:

OPENAI_API_KEY=your-openai-api-key
GEMINI_API_KEY=your-gemini-api-key

🚀 Quick Start

After connecting the server, ask your MCP client to create an image. For example:

Use Gemini model gemini-3.1-flash-image to create a 16:9 hero image for a coffee shop website. Save it as coffee-shop-hero in /Users/you/projects/coffee-shop/assets/images.

Use OpenAI model gpt-image-2 to create three square product icons and save them in /Users/you/projects/store/assets/images.

Every tool call must include absolute_output_directory. Relative paths are rejected, so the MCP server never depends on its process working directory.

Tools

Each item in images needs:

• prompt - What the image should contain

• filename - Base filename without an extension

📚 Notes

Output files

absolute_output_directory is required and must be an absolute directory path. The directory is created when it does not exist. For example:

/Users/you/projects/coffee-shop/assets/images

Files use this format:

{filename}-{provider}-{model}-{timestamp}.{ext}

Example:

coffee-shop-hero-gemini-gemini-3.1-flash-image-1784412779099.jpg

Provider options

• OpenAI supports model-specific settings such as size, quality, output format, and background.

• Gemini supports aspect ratio, image size, and JPEG output.

• Not every option works with every model. The provider's API rules still apply.

• If one image in a batch fails, successful images are still saved and the result lists each failure.

👋🏻 Contact