Architecture

System Overview

Skynet MCP Server uses a dual-database architecture for comprehensive AI memory:

┌─────────────────────────────────────────┐
│         MCP Client (AI Assistant)       │
└────────────────┬────────────────────────┘
                 │
                 │ Model Context Protocol
                 │
┌────────────────▼────────────────────────┐
│       Skynet MCP Server (FastMCP)       │
│  ┌──────────────────────────────────┐   │
│  │     15 Tools (5 Categories)      │   │
│  └──────────────────────────────────┘   │
└─────┬────────────────────────┬──────────┘
      │                        │
      │                        │
┌─────▼──────────┐      ┌──────▼─────────┐
│   Memgraph     │      │   ChromaDB     │
│  (Graph DB)    │      │  (Vector DB)   │
│                │      │                │
│ • Interactions │      │ • Memories     │
│ • Users        │      │ • Embeddings   │
│ • Topics       │      │ • Semantic     │
│ • Relations    │      │   Search       │
└────────────────┘      └────────────────┘

Components

Database Layer

Memgraph (Graph Database)

• Stores interactions between users and AI
• Tracks relationships (FOLLOWS, ABOUT, INITIATED)
• Maintains user profiles with interaction history
• Enables topic tracking and trend analysis

ChromaDB (Vector Database)

• Stores semantic memories as embeddings
• Enables similarity search for context retrieval
• Supports metadata filtering (type, user, importance)
• Handles preference storage and insights

Tool Categories

1. Infrastructure Tools (3)

• Container management via Docker API
• Network and volume orchestration
• Health monitoring

2. Database Tools (3)

• Direct Cypher query execution
• Vector search and indexing
• Document storage

3. Memory Tools (2)

• Semantic memory storage with metadata
• Context-aware memory retrieval
• Importance and confidence scoring

4. Interaction Tools (6)

• Full interaction lifecycle tracking
• User profile management
• Topic and entity extraction
• Analytics and insights

5. Cognitive Tools (1)

• Pre-processing workflow (skynet_think)

Data Flow

Storing an Interaction

User Input
    │
    ▼
skynet_think()
    │ (Extract context, topics)
    ▼
AI Processing
    │
    ▼
interaction_store()
    │
    ├──► Memgraph: Store interaction node
    │              Create relationships
    │              Link topics
    │
    └──► memory_store() (optional)
                   └──► ChromaDB: Persist semantic memory

Retrieving Context

User Query
    │
    ▼
memory_search()
    │ (Semantic search)
    ▼
ChromaDB: Find similar memories
    │
    ▼
interaction_getContext()
    │ (Graph traversal)
    ▼
Memgraph: Get recent interactions
    │
    ▼
Combine Results → Return Context

Design Patterns

1. Modular Architecture

• Each tool category in separate module
• Database clients isolated from business logic
• Type definitions centralized

2. Session Management

• Automatic session lifecycle for graph queries
• Proper cleanup in finally blocks
• Connection pooling

3. Type Safety

• Zod schemas for runtime validation
• TypeScript interfaces for compile-time safety
• Strict mode enabled

4. Error Handling

• Try-catch-finally for database operations
• Graceful degradation
• Meaningful error messages

Scalability Considerations

Horizontal Scaling

• Stateless server design
• Database clustering support (Memgraph, Chroma)
• Load balancer compatible

Performance

• Minified production builds (3.9MB)
• Indexed database queries
• Connection pooling
• Caching opportunities

Monitoring

• Container health checks
• Query performance metrics
• Memory usage tracking