Moondream

Moondream is an MCP server that connects Claude Desktop and other MCP clients to the Moondream vision language model for local and remote image analysis. Designed for developers, researchers, and automated visual workflows, it provides tools to extract descriptive information, query visual details, detect target objects, and return precise point coordinates from images. The server supports local file paths and remote URLs, and handles hardware acceleration automatically across Apple Silicon MPS, NVIDIA CUDA, and multi-core CPU fallbacks. Users can generate captions at varying detail levels, ask visual question answering queries in natural language, detect specific items with bounding box data, and point to items using coordinate mappings. Additionally, Moondream features batch processing capabilities, allowing clients to send multiple image assets simultaneously with concurrency controls. By running through the Model Context Protocol, the server allows language assistants to natively analyze diagrams, photos, and visual data without requiring external proprietary vision APIs.

Category: AI & LLM Tooling

Tags: captioning, computer-vision, image-analysis, object detection, vqa

Visit Moondream

How to install and configure Moondream

1. Ensure Python 3.10+ and PyTorch 2.0+ are installed on your machine. 2. Open your Claude Desktop configuration file at ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows). 3. Add the Moondream server under mcpServers using uvx: json { "mcpServers": { "moondream": { "command": "uvx", "args": ["moondream-mcp"], "env": { "MOONDREAM_DEVICE": "auto" } } } } 4. Alternatively, install via pip using pip install moondream-mcp and set the command to "moondream-mcp". 5. Save the configuration and restart Claude Desktop.

What you can do with Moondream

• Generating detailed, normal, or short image captions directly from local file paths or web URLs. * Answering natural language questions about specific visual elements, count of items, or scenes in images. * Detecting specified target objects within an image and returning their locations using bounding boxes. * Locating exact point coordinates for specific objects across user-provided visual media. * Processing batches of multiple images concurrently to caption or query sets of visual assets.

Key facts

• https://github.com/ColeMurray/moondream-mcp
• AI & LLM Tooling, Design, Media & Creative
• captioning, computer-vision, image-analysis, object detection, vqa

Part of MCP Servers

How do I install Moondream?

You can run Moondream without manual installation using uvx moondream-mcp, or install it directly from PyPI by running pip install moondream-mcp. It requires Python 3.10 or higher and PyTorch 2.0 or above.

What can Moondream do?

Moondream provides visual analysis tools including generating short or detailed captions, answering natural language questions about images, detecting objects with bounding boxes, extracting point coordinates, and running batch image operations across local files or URLs.

Which MCP clients work with Moondream?

Moondream works with any client that implements the Model Context Protocol, including Claude Desktop. You can configure it inside your client configuration JSON file by declaring the server command and environment variables.

What hardware is supported by Moondream?

The server automatically detects and optimizes for available hardware. It supports Apple Silicon via MPS on macOS, NVIDIA GPU acceleration via CUDA, and standard multi-core CPU processing as a fallback.

Is Moondream open source?

Yes, Moondream MCP is open source and licensed under the MIT License. The code is available on GitHub in the ColeMurray/moondream-mcp repository.