README.md

metadata

title: 🎥 Memvid MCP Server - Video-based AI Memory Storage
emoji: 🎥
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 5.31.0
app_file: app.py
pinned: true
license: mit
short_description: MCP server storing AI memories in MP4 videos with QR codes
models:
  - sentence-transformers/all-MiniLM-L6-v2
tags:
  - mcp-server-track
  - Agents-MCP-Hackathon
  - model-context-protocol
  - video-memory
  - semantic-search
  - ai-agents
  - memvid
  - faiss
  - huggingface

🎥 Memvid MCP Server

An advanced Model Context Protocol (MCP) server that stores AI conversation memories in MP4 video files using QR codes and semantic embeddings. Built for the Hugging Face Hackathon - MCP Server Track.

🚀 Live MCP Endpoint

https://eldarski-memvid-mcp-server.hf.space/gradio_api/mcp/sse

✨ Features

• 🎬 Video Memory Storage: Store text chunks in MP4 files with QR code encoding
• 🔍 Lightning-Fast Search: Semantic similarity search using FAISS embeddings
• 💬 Interactive Chat: Converse with your stored memories using AI
• ☁️ Cloud Integration: Automatic backup to HuggingFace datasets
• 🔧 24 MCP Tools: Comprehensive memory management via MCP protocol
• 🚀 91.7% Functional: Real working implementation with cloud storage

🎯 Quick Start

Add to MCP Client (Cursor, Claude Desktop, etc.)

{
  "mcpServers": {
    "memvid-server": {
      "url": "https://eldarski-memvid-mcp-server.hf.space/gradio_api/mcp/sse"
    }
  }
}

Basic Workflow

1. Store memories: store_memory(text, client_id)
2. Build video: build_memory_video(client_id, memory_name)
3. Search: search_memory(query, client_id, memory_name)
4. Chat: chat_with_memory(query, client_id, memory_name)

🔧 Available MCP Tools

Memory Operations

• store_memory - Store text chunks in video memory
• build_memory_video - Build MP4 memory from stored chunks
• search_memory - Semantic search in memory videos
• chat_with_memory - Interactive chat with memory
• list_memories - List all memories for a client
• get_memory_stats - Get memory usage statistics
• delete_memory - Delete specific memory videos
• store_document - Store document content in memory

HuggingFace Dataset Integration

• save_to_hf_dataset - Save client data to specific HF dataset
• load_from_hf_dataset - Load client data from HF dataset
• list_hf_datasets - List available HF datasets
• create_hf_dataset - Create new HF dataset
• get_storage_info - Get HF storage connection status
• backup_client_data - Backup to default HF dataset
• restore_client_data - Restore from default HF dataset

[## 🎬 Demo Video

[Link to demo video showing MCP server in action]](https://youtu.be/dnRH8NckB5A)

🏗️ How It Works

This MCP server uses the innovative memvid library to:

1. Encode text chunks into QR codes embedded in MP4 video frames
2. Generate semantic embeddings using sentence-transformers
3. Create FAISS indexes for lightning-fast similarity search
4. Enable AI chat with stored memories using context retrieval
5. Backup everything to HuggingFace datasets for persistence

Each client gets isolated storage with their own memory videos and embeddings.

📊 Test Results

• ✅ 91.7% Success Rate (22/24 tools working)
• ✅ Real Cloud Storage integration with HuggingFace
• ✅ PyTorch Compatibility solved for production deployment
• ✅ Memory Operations fully functional
• ✅ Search & Chat working with semantic embeddings

🛠️ Technical Stack

• Gradio - Web interface and MCP server
• Memvid - Video-based memory storage
• FAISS - Similarity search
• Sentence Transformers - Text embeddings
• HuggingFace - Cloud dataset storage

🏆 Hackathon Submission

Track: MCP Server / Tool
Tags: mcp-server-track
Status: Production-ready with 91.7% functionality
Innovation: First MCP server to use video files for AI memory storage

📄 License

MIT License - Feel free to use and modify!

🤝 Contributing

Built for the HuggingFace Hackathon. Contributions welcome!