Moondream
A vision language model for image analysis, including captioning, VQA, and object detection.
Moondream MCP Server
A FastMCP server for Moondream, an AI vision language model. This server provides image analysis capabilities including captioning, visual question answering, object detection, and visual pointing through the Model Context Protocol (MCP).
Features
- 🖼️ Image Captioning: Generate short, normal, or detailed captions for images
- ❓ Visual Question Answering: Ask natural language questions about images
- 🔍 Object Detection: Detect and locate specific objects with bounding boxes
- 📍 Visual Pointing: Get precise coordinates of objects in images
- 🔗 URL Support: Process images from both local files and remote URLs
- ⚡ Batch Processing: Analyze multiple images efficiently
- 🚀 Device Optimization: Automatic detection and optimization for CPU, CUDA, and MPS (Apple Silicon)
Installation
Prerequisites
- Python 3.10 or higher
- PyTorch 2.0+ (with appropriate device support)
Using uvx (Recommended for Claude Desktop)
# Run without installation
uvx moondream-mcp
# Or specify a specific version
uvx moondream-mcp==1.0.2
Install from PyPI
pip install moondream-mcp
Install from Source
git clone https://github.com/ColeMurray/moondream-mcp.git
cd moondream-mcp
pip install -e .
Development Installation
git clone https://github.com/ColeMurray/moondream-mcp.git
cd moondream-mcp
pip install -e ".[dev]"
Quick Start
Running the Server
# Using uvx (no installation needed)
uvx moondream-mcp
# Using pip-installed command
moondream-mcp
# Or run directly with Python
python -m moondream_mcp.server
Claude Desktop Integration
Add to your Claude Desktop configuration file:
macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json
Using uvx (Recommended)
{
"mcpServers": {
"moondream": {
"command": "uvx",
"args": ["moondream-mcp"],
"env": {
"MOONDREAM_DEVICE": "auto"
}
}
}
}
Using pip-installed command
{
"mcpServers": {
"moondream": {
"command": "moondream-mcp",
"env": {
"MOONDREAM_DEVICE": "auto"
}
}
}
}
Configuration
The server can be configured using environment variables:
Model Settings
MOONDREAM_MODEL_NAME
: Model name (default:vikhyatk/moondream2
)MOONDREAM_MODEL_REVISION
: Model revision (default:2025-01-09
)MOONDREAM_TRUST_REMOTE_CODE
: Trust remote code (default:true
)
Device Settings
MOONDREAM_DEVICE
: Force specific device (cpu
,cuda
,mps
, orauto
)
Image Processing
MOONDREAM_MAX_IMAGE_SIZE
: Maximum image dimensions (default:2048x2048
)MOONDREAM_MAX_FILE_SIZE_MB
: Maximum file size in MB (default:50
)
Performance
MOONDREAM_TIMEOUT_SECONDS
: Processing timeout (default:120
)MOONDREAM_MAX_CONCURRENT_REQUESTS
: Max concurrent requests (default:5
)MOONDREAM_ENABLE_STREAMING
: Enable streaming for captions (default:true
)MOONDREAM_MAX_BATCH_SIZE
: Maximum batch size for batch operations (default:10
)MOONDREAM_BATCH_CONCURRENCY
: Concurrent batch processing limit (default:3
)MOONDREAM_ENABLE_BATCH_PROGRESS
: Enable progress reporting for batch operations (default:true
)
Network (for URLs)
MOONDREAM_REQUEST_TIMEOUT_SECONDS
: HTTP request timeout (default:30
)MOONDREAM_MAX_REDIRECTS
: Maximum HTTP redirects (default:5
)MOONDREAM_USER_AGENT
: HTTP User-Agent header
Available Tools
1. caption_image
Generate captions for images.
Parameters:
image_path
(string): Path to image file or URLlength
(string): Caption length -"short"
,"normal"
, or"detailed"
stream
(boolean): Whether to stream caption generation
Example:
{
"image_path": "https://example.com/image.jpg",
"length": "detailed",
"stream": false
}
2. query_image
Ask questions about images.
Parameters:
image_path
(string): Path to image file or URLquestion
(string): Question to ask about the image
Example:
{
"image_path": "/path/to/image.jpg",
"question": "How many people are in this image?"
}
3. detect_objects
Detect specific objects in images.
Parameters:
image_path
(string): Path to image file or URLobject_name
(string): Name of object to detect
Example:
{
"image_path": "https://example.com/photo.jpg",
"object_name": "person"
}
4. point_objects
Get coordinates of objects in images.
Parameters:
image_path
(string): Path to image file or URLobject_name
(string): Name of object to locate
Example:
{
"image_path": "/path/to/image.jpg",
"object_name": "car"
}
5. analyze_image
Multi-purpose image analysis tool.
Parameters:
image_path
(string): Path to image file or URLoperation
(string): Operation type ("caption"
,"query"
,"detect"
,"point"
)parameters
(string): JSON string with operation-specific parameters
Example:
{
"image_path": "https://example.com/image.jpg",
"operation": "query",
"parameters": "{\"question\": \"What is the weather like?\"}"
}
6. batch_analyze_images
Process multiple images in batch.
Parameters:
image_paths
(string): JSON array of image pathsoperation
(string): Operation to perform on all imagesparameters
(string): JSON string with operation-specific parameters
Example:
{
"image_paths": "[\"image1.jpg\", \"image2.jpg\"]",
"operation": "caption",
"parameters": "{\"length\": \"short\"}"
}
Usage Examples
Basic Image Captioning
# Using the caption_image tool
result = await caption_image(
image_path="https://example.com/sunset.jpg",
length="detailed"
)
Visual Question Answering
# Ask about image content
result = await query_image(
image_path="/path/to/family_photo.jpg",
question="How many children are in this photo?"
)
Object Detection
# Detect faces in an image
result = await detect_objects(
image_path="https://example.com/group_photo.jpg",
object_name="face"
)
Batch Processing
# Process multiple images
result = await batch_analyze_images(
image_paths='["img1.jpg", "img2.jpg", "img3.jpg"]',
operation="caption",
parameters='{"length": "normal"}'
)
Device Support
The server automatically detects and optimizes for available hardware:
Apple Silicon (MPS)
- Optimal performance on M1/M2/M3 Macs
- Automatic memory management
- Native acceleration
NVIDIA CUDA
- GPU acceleration for NVIDIA cards
- Automatic CUDA memory management
- Mixed precision support
CPU Fallback
- Works on any system
- Optimized for multi-core processing
- Lower memory requirements
Error Handling
The server provides detailed error information:
{
"success": false,
"error_message": "Image file not found: /path/to/missing.jpg",
"error_code": "IMAGE_PROCESSING_ERROR",
"processing_time_ms": 15.2
}
Common error codes:
MODEL_LOAD_ERROR
: Issues loading the Moondream modelIMAGE_PROCESSING_ERROR
: Problems with image files or URLsINFERENCE_ERROR
: Model inference failuresINVALID_REQUEST
: Invalid parameters or requests
Performance Tips
- Use appropriate image sizes: Resize large images before processing
- Batch processing: Use
batch_analyze_images
for multiple images - Device optimization: Let the server auto-detect the best device
- Concurrent requests: Adjust
MOONDREAM_MAX_CONCURRENT_REQUESTS
based on your hardware - Memory management: Monitor memory usage, especially with large images
Troubleshooting
Model Loading Issues
# Check PyTorch installation
python -c "import torch; print(torch.__version__)"
# Check device availability
python -c "import torch; print(f'CUDA: {torch.cuda.is_available()}, MPS: {torch.backends.mps.is_available()}')"
Memory Issues
- Reduce
MOONDREAM_MAX_IMAGE_SIZE
- Lower
MOONDREAM_MAX_CONCURRENT_REQUESTS
- Use CPU instead of GPU for large images
Network Issues
- Check firewall settings for URL access
- Increase
MOONDREAM_REQUEST_TIMEOUT_SECONDS
- Verify SSL certificates for HTTPS URLs
Development
Running Tests
pytest tests/
Code Quality
# Format code
black src/ tests/
# Sort imports
isort src/ tests/
# Type checking
mypy src/
Contributing
- Fork the repository
- Create a feature branch
- Make your changes
- Add tests
- Run quality checks
- Submit a pull request
License
This project is licensed under the MIT License. See LICENSE for details.
Acknowledgments
- Moondream - The amazing vision language model
- FastMCP - The MCP server framework
- Model Context Protocol - The protocol specification
Support
Note: This server requires downloading the Moondream model on first use, which may take some time depending on your internet connection.
Related Servers
plugged.in MCP Proxy Server
A middleware that aggregates multiple Model Context Protocol (MCP) servers into a single unified interface.
Postman API
An MCP server for interacting with the Postman API, requiring an API key.
Ref
Up-to-date documentation for your coding agent. Covers 1000s of public repos and sites. Built by ref.tools
LogAI MCP Server
An MCP server for log analysis using the LogAI framework, with optional Grafana and GitHub integrations.
ZenML
Interact with your MLOps and LLMOps pipelines through your ZenML MCP server
Remote MCP Server (Authless)
An example of a remote MCP server deployable on Cloudflare Workers, without authentication.
Gemini Imagen 3.0
Generate high-quality images using Google's Imagen 3.0 model via the Gemini API.
Ilograph MCP Server
Create and validate Ilograph diagrams with access to documentation and guidance.
Unity-MCP
A bridge between the Unity game engine and AI assistants using the Model Context Protocol (MCP).
OpenRPC MCP Server
Provides JSON-RPC functionality through the OpenRPC specification.