README.md

metadata

title: ProjectMemory
emoji: ⚡
colorFrom: red
colorTo: indigo
sdk: docker
app_port: 7860
pinned: false
license: mit
short_description: Semantic, shared AI project memory.
tags:
  - building-mcp-track-enterprise

🎯 Track 1: Building MCP - Enterprise Category

Project Memory is a multi-user, multi-project AI memory system powered by MCP (Model Context Protocol). It creates shared project memory where every action gets logged and becomes searchable via semantic search and AI chat.

🚀 What We Built

An MCP server that extends LLM capabilities for enterprise teams by:

• Persistent Project Memory: Every task completion generates AI documentation that becomes searchable knowledge
• Semantic Search: Vector-based memory retrieval across all project activities
• MCP Tool Integration: Exposes project management capabilities as MCP tools
• Multi-User Collaboration: Teams can share and search collective knowledge

🛠️ MCP Tools Exposed

Our MCP server provides these tools:

• create_project: Initialize a new project workspace
• list_projects: View all available projects
• join_project: Join an existing project
• list_tasks: Get project tasks with status
• complete_task: Mark task as done with AI-generated documentation
• memory_search: Semantic search across project history
• list_activity: View project activity feed

📹 Demo Video

Watch our 3-minute demo showing MCP integration with Claude Desktop (link to be added)

🏗️ Architecture

┌─────────────────┐     ┌─────────────────┐
│   Web Frontend  │────▶│  FastAPI Backend │
│     (React)     │     │   (MCP Client)   │
└─────────────────┘     └─────────────────┘
                               │
                               ▼
                        ┌─────────────────┐
                        │   MCP Server    │
                        │  (TypeScript)   │
                        └─────────────────┘
                               │
                               ▼
                        ┌─────────────────┐
                        │  SQLite + Vec   │
                        │   (Embeddings)  │
                        └─────────────────┘

💡 Key Features

1. Task Completion Pipeline: Transforms user work into searchable documentation
2. Vector Search: Semantic retrieval using sqlite-vec embeddings
3. Chat Interface: Natural language queries using MCP tools
4. Activity Feed: Real-time project activity tracking
5. Multi-Project Support: Manage multiple projects with isolated memory

🔧 Technical Stack

• MCP Server: TypeScript with @modelcontextprotocol/sdk
• Backend: FastAPI (Python) as MCP client
• Frontend: React + Vite + Tailwind CSS
• Database: SQLite with sqlite-vec for embeddings
• AI: Google Generative AI (Gemini) for documentation generation
• Deployment: Docker container for Hugging Face Spaces

🎮 How to Use

1. Create or Join a Project: Start by creating a new project or joining an existing one
2. Complete Tasks: Mark tasks as done and provide context about your work
3. AI Documentation: System automatically generates searchable documentation
4. Search Memory: Use semantic search to find any past work or decision
5. Chat with Memory: Ask questions about project history using natural language

🚢 Deployment

This Space runs as a Docker container combining:

• FastAPI backend serving as MCP client
• React frontend for user interface
• MCP server handling all tool operations
• SQLite database with vector search capabilities

🔐 Environment Variables

Configure in Space settings:

• GOOGLE_API_KEY: For Gemini AI integration
• DATABASE_URL: (Optional) Custom database connection

👥 Team

Add team member names here

📝 License

MIT License - See LICENSE file for details

🔗 Links

• GitHub Repository
• MCP Documentation
• Hackathon Page