jenreh/image.serv

Image Generation MCP Server

A FastMCP server for generating and editing images using OpenAI's gpt-image-1 and Azure FLUX.1-Kontext-pro models. Deploy as an MCP tool or use REST endpoints for flexible integration with AI assistants and applications.

Features

• Text-to-Image Generation: Create images from natural language prompts using multiple AI models
• Image Editing & Inpainting: Edit existing images with text prompts and optional masks (gpt-image-1)
• Multiple Formats: Output as PNG, JPEG, or WEBP with customizable quality
• Flexible Response Formats: Receive results as MCP Images, Markdown, or Microsoft Adaptive Cards
• Prompt Enhancement: Auto-refine prompts via LLM for better results
• Dual Protocol: Access via MCP protocol for AI assistants or REST API for direct integration

Quick Start

Prerequisites

• Python 3.12 or later
• uv package manager
• API keys:
  • REQUIRED: Azure OpenAI API key and endpoint (for gpt-image-1 model)
  • OPTIONAL: Google AI API key (for additional image generation capabilities)

Installation

1. Clone and navigate to the project:

git clone <repository>
cd image.serv

1. Install dependencies:

uv sync

1. Configure environment variables:

cp .env.example .env
# Edit .env with your actual values

Or set them directly:

# REQUIRED: Azure OpenAI API Configuration
export OPENAI_API_KEY="your-azure-openai-api-key" # pragma: allowlist secret
export OPENAI_BASE_URL="https://your-resource-name.openai.azure.com"

# OPTIONAL: Google AI Configuration
export GOOGLE_API_KEY="your-google-api-key" # pragma: allowlist secret

# OPTIONAL: Backend server URL for image download URLs
export BACKEND_SERVER="http://localhost:8000"

# OPTIONAL: Temporary image storage directory
export TMP_PATH="./images"

# OPTIONAL: Default response format (image|markdown|adaptive_card)
export DEFAULT_RESPONSE_FORMAT="markdown"

# OPTIONAL: Logging level (DEBUG|INFO|WARNING|ERROR|CRITICAL)
export LOG_LEVEL="INFO"

# OPTIONAL: Server port
export PORT="8000"

Usage

Running the Server

Start the server for integration with Claude Desktop or other MCP clients:

uv run python -m server.server

The server will be available at http://localhost:8000 with endpoints:

• POST /api/v1/generate_image - Generate images
• POST /api/v1/edit_image - Edit existing images
• /api/docs - OpenAPI Documenation

MCP Configuration

{
    "servers": {
        "image-generation": {
            "type": "http",
            "url": "http://localhost:8000/mcp/v1",
            "gallery": true
        }
    }
}

Tools & API

generate_image

Create images from text descriptions.

Parameters:

Example:

generate_image(
    prompt="A serene mountain landscape at sunset with golden light reflecting off a lake",
    size="1536x1024",
    output_format="png",
    enhance_prompt=True,
    response_format="markdown"
)

edit_image

Edit existing images with text prompts and optional masks for inpainting.

Parameters:

Example:

edit_image(
    prompt="Add a vibrant rainbow across the sky",
    image_paths=["https://example.com/landscape.jpg"],
    mask_path="https://example.com/sky_mask.png",
    output_format="png",
    response_format="markdown"
)

Image Input Formats

Supported input methods for image_paths parameter:

• HTTP/HTTPS URLs: https://example.com/image.jpg
• Local file paths: /path/to/image.png
• Base64 data URLs: data:image/png;base64,iVBORw0KG...

Inpainting with Masks

For precise control over edits, use mask images:

1. Create a PNG image with alpha transparency
2. Transparent areas (alpha=0) mark regions to edit
3. Opaque areas remain unchanged
4. Mask dimensions must match the input image

Architecture

The server follows a clean, layered architecture:

MCP Client / REST Client
    ↓
FastMCP Server / FastAPI Routes
    ↓
Image Service Layer
    ↓
Image Generators (OpenAI, Google)
    ├── Prompt Enhancer
    ├── Image Processor
    └── Image Loader

Key Components

• mcp_server.py - FastMCP tool definitions and server setup
• server.py - Unified MCP + REST API server
• api/routes.py - FastAPI REST endpoints
• backend/image_service.py - Core business logic
• backend/generators/ - AI provider implementations
  • openai.py - OpenAI gpt-image-1 integration

Development

Running Tests

make test

Run with coverage:

make test-coverage

Code Quality

Format and lint code:

make format
make check

Available Commands

make help

Response Formats

All endpoints support three response formats via the response_format parameter:

Image Format

Returns raw image data suitable for display or processing.

Markdown Format

Returns formatted markdown with embedded base64 images:

# Generated Image

![Generated Image](http://localhost:8000/_uploads/...)

Adaptive Card Format

Returns Microsoft Adaptive Card JSON for Teams or other platforms:

{
  "type": "AdaptiveCard",
  "version": "1.4",
  "body": [
    {
      "type": "Image",
      "url": "http://localhost:8000/_uploads/..."
    }
  ]
}

Error Handling

The API returns standard HTTP status codes:

• 200 OK - Successful generation/editing
• 400 Bad Request - Invalid parameters
• 422 Unprocessable Entity - Validation error
• 500 Internal Server Error - API failure

Error responses include detailed messages:

{
  "status": "error",
  "error": "Description of what went wrong",
  "metadata": {
    "timestamp": "2025-01-15T10:30:00Z"
  }
}

Environment Variables

License

This project is licensed under the MIT License - see for details.