MCP Memory Server

# Generic Memory System Design ## Overview Replace the rigid `global/learned/agent` memory types with a flexible, user-defined collection system. ## Current Problems - ❌ Only 3 hardcoded memory types: `global`, `learned`, `agent` - ❌ Fixed collection names and structures - ❌ Rigid API methods tied to specific types - ❌ No way to create custom collections for different projects/contexts - ❌ Memory browser locked to predefined types ## Proposed Generic System ### 1. Flexible Collections ```typescript interface Collection { name: string; // "project-alpha", "meeting-notes", "research" description?: string; // User-provided description tags: string[]; // ["work", "personal", "code", "docs"] metadata: { created_at: string; created_by: string; permissions: { read: string[]; // Users/agents who can read write: string[]; // Users/agents who can write admin: string[]; // Users/agents who can manage }; category?: string; // "documentation", "code", "notes", "research" project?: string; // Associated project/context retention?: number; // Days to keep (optional TTL) }; stats: { document_count: number; last_updated: string; size_bytes: number; }; } ``` ### 2. Generic Memory Operations ```python # Create collections dynamically create_collection(name="project-alpha", description="Alpha project notes", tags=["work", "development"]) # Add content to any collection add_memory(collection="project-alpha", content="...", metadata={...}) # Search across multiple collections search_memory(query="bug fix", collections=["project-alpha", "troubleshooting"], limit=10) # Manage collections list_collections(filter_by_tags=["work"], owned_by="user123") update_collection(name="project-alpha", new_tags=["work", "completed"]) delete_collection(name="old-project", confirm=True) ``` ### 3. Backward Compatibility Migration - `global_memory` → `shared-knowledge` collection with tag ["global", "shared"] - `learned_memory` → `patterns-insights` collection with tag ["learned", "patterns"] - `agent_specific_memory_X` → `agent-X-context` collection with tag ["agent", "personal"] ### 4. Enhanced Memory Browser UI - **Collection Browser**: Tree view of all collections with metadata - **Dynamic Creation**: "New Collection" button with form - **Tag Filtering**: Filter collections by tags, owner, project - **Bulk Operations**: Move content between collections, merge collections - **Collection Templates**: Quick create from templates (project, research, etc.) ### 5. Smart Collection Suggestions Based on content analysis, suggest appropriate collection: - Code snippets → "code-library" - Meeting notes → "meetings-2024" - Bug reports → "troubleshooting" - Documentation → "project-docs" ## Implementation Plan ### Phase 1: Core Infrastructure 1. **New Collection Model**: Replace hardcoded types with dynamic collections 2. **Collection Manager**: CRUD operations for collections 3. **Migration Script**: Convert existing data to new format ### Phase 2: API Updates 1. **Generic Memory API**: Replace type-specific methods 2. **Flexible Query Engine**: Search across any collections 3. **Permission System**: Collection-level access control ### Phase 3: UI Enhancement 1. **Dynamic Collection Browser**: Show all collections, not just 3 types 2. **Collection Management UI**: Create, edit, delete collections 3. **Smart Content Organization**: Suggest collections based on content ## Benefits - ✅ **Infinite Flexibility**: Create any collection structure users need - ✅ **Better Organization**: Group content by project, team, topic, etc. - ✅ **User-Centric**: Let users define their own memory organization - ✅ **Scalable**: No limits on number or types of collections - ✅ **Professional**: Proper permissions, metadata, lifecycle management