MARM MCP Server

Universal MCP Memory Intelligence

MARM (Memory Accurate Response Mode) provides persistent, cross-session memory for AI agents through semantic search, structured logging, and intelligent context management. Stop fighting your AI's memory—give it a persistent layer that remembers decisions, code, and conversations across all your workflows.

Key Features

AI-Powered Memory

• Semantic search using vector embeddings (all-MiniLM-L6-v2)
• Auto-classification of content (code, project, book, general)
• Persistent cross-session memory with SQLite + WAL mode
• Smart recall with context-aware fallbacks

Multi-Transport Support

• HTTP transport with mock OAuth for local dev
• STDIO transport for CLI tools and orchestration
• WebSocket transport (beta) for real-time communication

18 Complete MCP Tools

• marm_smart_recall - AI-powered semantic similarity search across all memories
• marm_contextual_log - Intelligent auto-classifying memory storage
• marm_start / marm_refresh - Session management
• marm_log_session / marm_log_entry / marm_log_show / marm_log_delete - Structured logging
• marm_summary - Context-aware summaries with intelligent truncation
• marm_context_bridge - Smart context bridging for workflow transitions
• marm_notebook_add / marm_notebook_use / marm_notebook_show / marm_notebook_delete / marm_notebook_clear / marm_notebook_status - Notebook management
• marm_current_context - Auto date/time for logs
• marm_system_info / marm_reload_docs - System utilities

Architecture

• FastAPI 0.115.4 + FastAPI-MCP 0.4.0
• SQLite with WAL mode + custom connection pooling
• Sentence Transformers for semantic embeddings
• IP-based rate limiting with tiered throttling
• MCP response size compliance (1MB limit)
• Docker containerization with health monitoring

Usage

Just talk naturally to your AI agent:

• "Claude, log this session as 'Project Alpha' and add this conversation as 'database design discussion'"
• "Remember this code snippet in your notebook for later"
• "Search for what we discussed about authentication yesterday"

The AI agent automatically uses the appropriate MARM tools. Manual tool access available for power users.

Production Features

• Custom SQLite connection pool (thread-safe, configurable)
• Lazy loading of semantic models
• Intelligent caching with cleanup cycles
• Rate limiting: 60 req/min default, 20 req/min for memory-heavy ops
• Usage analytics and business intelligence tracking