VisionCraft MCP Server

Website

GitHub

📺 VisionCraft MCP Demo

❌ Without VisionCraft MCP

AI assistants struggle with computer vision and Gen-AI knowledge:

❌ Limited understanding of cutting-edge computer vision algorithms
❌ Outdated information on models released after training cutoff
❌ Generic answers that lack depth and technical accuracy
❌ No awareness of latest frameworks and tools

✅ With VisionCraft MCP

VisionCraft MCP delivers up-to-date, specialized computer vision and Gen-AI knowledge directly to Claude and other AI assistants.

✅ Specialized knowledge on state-of-the-art computer vision and Gen-AI
✅ Up-to-date information on the newest models and techniques
✅ Detailed, technically accurate responses to complex queries
✅ Full awareness of latest frameworks, tools, and best practices

Add use visioncraft mcp to your prompt in Claude:

Give me the code for OpenAI Agents SDK. Use visioncraft MCP

VisionCraft MCP provides:

🔍 Specialized Knowledge: Access to state-of-the-art computer vision and GenAI information
🦅 RAVEN Engine: Advanced knowledge processing system that delivers precise, context-aware information
🚀 Latest Algorithms: Stay updated with cutting-edge models and techniques
💡 Expert Guidance: Get implementation advice from computer vision and GenAI specialists

🛠️ Getting Started

Requirements

Node.js >= v18.0.0
Claude Desktop or another MCP-compatible client

Install in MCP-compatible Clients (Claude Desktop, Cursor, Windsurf, VS Code/Cline)

Add the following configuration to your respective MCP settings file. For Claude Desktop, this is typically claude_desktop_config.json. For VS Code, you might use a general MCP configuration file or specific extension settings. For Windsurf, Cursor, and other Cline-like tools, refer to their documentation for the MCP configuration location.

Note: While the command and arguments are generally the same, the top-level key for MCP servers might vary (e.g., mcpServers or servers). The example below uses mcpServers. Adjust if your specific client requires a different key (like servers for some VS Code setups).

{
  "mcpServers": {
    "visioncraft": {
      "command": "npx",
      "args": ["-y", "visioncraft-mcp@latest"]
    }
  }
}

Demo: VisionCraft MCP in Action

Windsurf Integration

<img src="demos/windsurf_demo.gif" alt="VisionCraft MCP in Windsurf" width="100%"/> VisionCraft MCP in Windsurf

VS Code Cline Integration

<img src="demos/cline_demo.gif" alt="VisionCraft MCP in Cline" width="100%"/> VisionCraft MCP in Cline

🧠 Knowledge Base Coverage

VisionCraft MCP provides expertise in Computer Vision and GenAI including:

Object Detection

YOLOv12 and earlier versions
RT-DETR and DAB-DETR transformers
CenterNet and keypoint-based approaches
Ultralytics frameworks
Roboflow supervision library

Segmentation

SAM/SAM2 (Segment Anything Model)
FastSAM, MobileSAM, EfficientSAM
Domain adaptation techniques
3D instance segmentation

3D Vision

Point cloud detection frameworks
Feature extraction for 3D objects
Multi-view geometry transformers
Real-time 3D detection systems

Vision-Language Models

Florence-2, Qwen2.5-VL
Spatial language models
Foundation vision-language models

Specialized Computer Vision

Occupancy and velocity transformers
Event-based vision systems
Depth estimation techniques
Synthetic motion capture

Agentic Frameworks

OpenAI Agents SDK
CrewAI

More libraries coming soon.

🔍 Available Tools

vision-query: Query the VisionCraft knowledge base for computer vision and GenAI information
- query (required): The search query about computer vision topics

🗺️ Roadmap

VisionCraft MCP is actively evolving. We are currently in a free testing phase as we build out our core functionalities and gather user feedback. Our goal is to provide a powerful tool for developers working with Computer Vision and GenAI.

Currently Available:

[x] Access to a foundational knowledge base covering key Computer Vision and GenAI topics (Object Detection, Segmentation, 3D Vision, Vision-Language Models, Agentic Frameworks like OpenAI Agents SDK & CrewAI).
[x] vision-query tool for direct information retrieval via MCP.
[x] Easy integration with MCP-compatible clients (Claude Desktop, Cursor, Windsurf, VS Code/Cline).

In Progress / Near Term:

[ ] Expanded Knowledge Domains: Broadening coverage to include popular general programming frameworks, RAG-specific frameworks (e.g., LangChain, LlamaIndex), and other developer tools.
[ ] Enhanced Client Integration: Deeper and more seamless integration with Claude Desktop, Cursor, and Windsurf, including client-specific optimizations and improved performance.
[ ] User Accounts & Personalized Experience: Implementing mechanisms for user accounts and managing access, laying groundwork for personalized experiences.
[ ] Performance Optimizations: Continuously improving response times and resource efficiency of the MCP server.

Future Goals:

[ ] Comprehensive Framework Support: Aiming to cover a vast array of programming languages, libraries, and developer ecosystems.
[ ] Advanced RAG & Agentic Workflow Support: Offering more sophisticated Retrieval Augmented Generation techniques and tools, along with support for building and understanding agentic workflows through the MCP.
[ ] Interactive Code Generation & Assistance: Assisting with boilerplate code, debugging suggestions, and framework-specific snippets based on the expanded knowledge.
[ ] Custom Knowledge Base Integration: Allowing users or enterprises to connect their own private knowledge bases for personalized assistance.

Note:

VisionCraft MCP is currently free during development. We plan to introduce paid tiers in the future to support ongoing maintenance and expansion. Early users providing feedback will be valued as we shape our offerings.

📄 License

MIT License

🚀 About VisionCraft

VisionCraft helps developers build computer vision and GenAI applications using the latest and state-of-the-art algorithms, models, and frameworks. Our mission is to make advanced computer vision and GenAI technology accessible to developers of all skill levels.

Visit VisionCraft to learn more.

Made with ❤️ by Augmented AI