Collective Brain MCP Server

# AtlasMCP - Your AI Second Brain **Universal AI Memory System** - AtlasMCP allows any AI user to have a persistent, intelligent memory across all platforms, breaking down the barriers between AI tools and enabling true collaborative intelligence. ## The Problem We Solve Today, each AI platform has its own closed memory system, and AIs can't share a common context. Users lose valuable information when switching between tools, and teams can't leverage collective knowledge effectively. ## Our Solution AtlasMCP is what allows any AI user today to have a second brain. Every time they interact with an AI that has AtlasMCP enabled, all of their important data is automatically indexed into a vector database that acts like their extended memory. This means that whether they're chatting on Mistral, coding with Cursors, or using any other AI tool, they now have one unified database containing all their context and history. With AtlasMCP, the barrier between AI platforms is gone. It's a real breakthrough in context management, giving every user a persistent, intelligent memory across platforms. ## Key Features ### Personal Second Brain - **Automatic Indexing**: All important data from AI interactions is automatically stored - **Cross-Platform Memory**: Unified database accessible from any AI tool - **Semantic Search**: Find information using natural language queries - **Persistent Context**: Never lose important conversations or decisions ### Team Collaboration - **Shared Knowledge Base**: Teams can access collective memory through our web app - **Multi-Tenant Architecture**: Secure isolation between teams and organizations - **Real-Time Analytics**: Track team knowledge patterns and engagement - **Granular Permissions**: Control access with private/team/public visibility levels ### MCP Tools Available 1. **`add_memory`** - Store intelligent memories - Content, category, tags, visibility (private/team/public) - Automatic importance detection and similarity matching - Qdrant vector database integration - Multi-tenant authentication via Supabase 2. **`search_memories`** - Advanced semantic search - Content similarity search with embeddings - Filters by category, visibility, and team - Relevance + confidence scoring - Granular permission respect 3. **`get_team_insights`** - Real-time team analytics - Top categories and most used tags - Most active contributors - Most accessed memories - Team engagement metrics 4. **`delete_memory`** - Memory management - Secure deletion with permission verification - Automatic reference cleanup 5. **`list_memories`** - Knowledge base exploration - Paginated team memory listing - Filtering by user and category ## Business Value ### The Second Major Benefit: Team Collaboration We've built a standard database on Supabase with team workspaces. Through our web app, you can create a team and invite members. Everyone in that team gets access to the same collective memory. Imagine a startup: the CTO and CEO can both access not only the technical documentation, but also business-critical knowledge like the status of fundraising, VC conversations, and client feedback. Instead of scattered, personal silos of memory, the entire team gains a shared, living knowledge base, accessible from any AI tool connected to AtlasMCP. This is a revolution for how teams work with AI: personal second brains, and collective brains for organizations. ## Technical Architecture ``` ┌─────────────────┐ ┌──────────────────┐ ┌─────────────────┐ │ Mistral Chat │ │ Cursor IDE │ │ Other AI │ │ (CEO Alice) │ │ (CTO Bob) │ │ (CS Charlie) │ └─────────┬───────┘ └─────────┬────────┘ └─────────┬───────┘ │ │ │ └──────────────────────┼───────────────────────┘ │ ┌─────────────▼─────────────┐ │ AtlasMCP Server │ │ - Universal Memory │ │ - Multi-tenant Auth │ │ - Semantic Search │ │ - Team Collaboration │ └─────────────┬─────────────┘ │ ┌─────────────▼─────────────┐ │ Storage Layer │ │ ├─ Qdrant (Vectors) │ │ ├─ Supabase (Auth/DB) │ │ └─ Memory (Fallback) │ └───────────────────────────┘ ``` ### Technology Stack: - **Backend**: Python 3.13 + FastMCP + FastAPI - **Vector DB**: Qdrant (cloud + local fallback) - **Auth/DB**: Supabase (PostgreSQL + RLS) - **Deployment**: AWS Lambda + Alpic - **Frontend**: Next.js + TypeScript (webapp) ## Installation & Configuration ### 1. **Local Installation** ```bash # Clone the project git clone https://github.com/bparpette/MistralHackathon.git cd MistralHackathon # Install dependencies uv sync # Configure environment cp example.env config.env # Edit config.env with your API keys ``` ### 2. **Required Configuration** ```bash # Required environment variables QDRANT_URL=https://your-cluster.qdrant.io:6333 QDRANT_API_KEY=your_qdrant_api_key SUPABASE_URL=https://your-project.supabase.co SUPABASE_SERVICE_ROLE_KEY=your_supabase_service_key ``` ### 3. **Automatic Deployment** - **Production**: Automatic deployment via Alpic on every commit to `main` - **URL**: https://mistralhackathonmcp-ee61017d.alpic.live/ - **Configuration**: API keys configured directly on the platform ### 4. **Supabase Database** ```sql -- Execute the complete schema in Supabase SQL editor -- See webapp/supabase-schema.sql for the complete schema ``` ## Demo Scenario ### "Startup AI - Critical Bug Resolved in 45 min instead of 2h" ```bash # Launch local server uv run main.py # Test MCP tools # Use the 5 tools: add_memory, search_memories, get_team_insights, delete_memory, list_memories ``` **Measured Impact:** 75% reduction in resolution time thanks to instant information sharing! ### Demo Workflow: 1. **CS receives complaint** → Stores in collective memory 2. **CEO searches context** → Immediately finds business impact (500k€/year) 3. **CTO debugs** → Sees maximum priority instantly 4. **CTO resolves** → Documents the solution 5. **CS reassures client** → Has all technical details ## Use Cases ### 1. **Critical Problem Resolution** - **Before**: 2h of research + coordination - **After**: 45 min of direct resolution - **Gain**: 75% time reduction ### 2. **Informed Decision Making** - Documented and traceable decisions - Historical context instantly accessible - Collaborative information validation ### 3. **Accelerated Onboarding** - New members access complete history - Preserved and organized knowledge - Best practices automatically shared ### 4. **Team Analytics** - Expert identification by domain - Usage and engagement patterns - Internal process optimization ## Advanced Configuration ### Environment Variables: ```bash # Qdrant (Vector Database) QDRANT_URL=https://your-cluster.qdrant.io:6333 QDRANT_API_KEY=your_qdrant_api_key QDRANT_ENABLED=true # Supabase (Auth & Database) SUPABASE_URL=https://your-project.supabase.co SUPABASE_SERVICE_ROLE_KEY=your_supabase_service_key # MCP Server MCP_SERVER_PORT=3000 MCP_SERVER_DEBUG=false # AWS Lambda (Production) AWS_LAMBDA_FUNCTION_NAME=your_function_name AWS_EXECUTION_ENV=AWS_Lambda_python3.13 ``` ### Permission System: - **`private`** - Only the creator can see - **`team`** - All team members - **`public`** - Accessible to all (with authentication) ### Multi-tenant: - Complete isolation by team via `team_token` - Authentication via unique `user_token` - RLS (Row Level Security) on all tables ## Roadmap & Status ### Phase 1 (MVP) - **COMPLETED** - [x] Multi-tenant MCP server operational - [x] Memory storage with Qdrant - [x] Advanced semantic search - [x] Granular permission system - [x] Real-time team analytics - [x] Supabase authentication + RLS - [x] Automatic AWS Lambda deployment - [x] Next.js + TypeScript webapp ### Phase 2 (Production) - **IN PROGRESS** - [x] Operational Qdrant cloud integration - [x] Robust memory fallback - [x] Automated Alpic deployment - [ ] Complete web dashboard - [ ] Public REST API - [ ] Monitoring and alerts ### Phase 3 (Evolution) - **PLANNED** - [ ] Automatic knowledge graph - [ ] Advanced Mistral embeddings - [ ] Multi-language (EN/FR/ES) - [ ] Voice notes and transcription - [ ] AI predictive insights - [ ] Third-party integrations (Slack, Teams) ## Competitive Advantages 1. **First MCP collective memory system** - Unique technical innovation 2. **Native multi-tenant** - Complete team isolation with RLS 3. **Granular permissions** - Fine access control (private/team/public) 4. **Advanced semantic search** - Finds information even with different words 5. **Real-time analytics** - Insights on team activity and engagement 6. **Optimized performance** - Robust fallback + Lambda deployment 7. **MCP ecosystem** - Compatible with all MCP clients 8. **Measurable impact** - 75% reduction in resolution time ## Deployment & URLs ### Production: - **MCP Server**: https://mistralhackathonmcp-ee61017d.alpic.live/ - **Repository**: https://github.com/bparpette/MistralHackathon - **Webapp**: Next.js + TypeScript (folder `webapp/`) ### MCP Configuration: ```json { "mcpServers": { "atlasmcp": { "url": "https://mistralhackathonmcp-ee61017d.alpic.live/", "headers": { "Authorization": "Bearer user_d8a7996df3c777e9ac2914ef16d5b501" } } } } ``` ## Project Structure ``` MistralHackathon/ ├── main.py # Main MCP server ├── pyproject.toml # Python dependencies ├── config.env # Production configuration ├── example.env # Configuration template ├── webapp/ # Next.js frontend │ ├── src/app/ # Pages and API routes │ ├── supabase-schema.sql # Complete DB schema │ └── package.json # Frontend dependencies ├── doc/ # Documentation │ ├── HACKATHON_SUMMARY.md │ ├── DEVELOPMENT_SUMMARY.md │ └── archi.md ├── qdrant_storage/ # Local Qdrant data └── LEGACY/ # Previous versions ``` ## Team & Contribution This project was developed during the **Mistral AI MCP Hackathon 2025** by the AtlasMCP team. ### Developers: **Baptiste Parpette** - LinkedIn: [baptiste-parpette](https://www.linkedin.com/in/baptiste-parpette/) - GitHub: [@Bparpette](https://github.com/bparpette) **Henri d'Aboville** - LinkedIn: [henri-d-52bb1a383](https://www.linkedin.com/in/henri-d-52bb1a383/) ## License MIT License - See LICENSE file for more details. --- **AtlasMCP - Transform your team into an intelligent collective brain!**