Memento MCP: A Knowledge Graph Memory System for LLMs

Memento MCP is a knowledge graph memory system designed to provide Large Language Models (LLMs) with persistent, adaptive, and long-term ontological memory. It targets LLM clients supporting the Model Context Protocol (MCP), such as Claude Desktop and Cursor, offering benefits like semantic retrieval, contextual recall, and temporal awareness. This system enhances LLM capabilities by enabling resilient and adaptive memory management.

How It Works

The system utilizes Neo4j as a unified backend, consolidating both graph storage and vector search functionalities. Entities and relations form the knowledge graph, with entities possessing unique names, types, observations, and vector embeddings for semantic search. Relations connect entities with properties like strength, confidence, and metadata, incorporating temporal awareness through version history. This approach leverages Neo4j's native graph operations and integrated vector search for a simplified, high-performance architecture.

Quick Start & Requirements

• Primary Install/Run: Recommended for MCP clients: npx -y @gannonh/memento-mcp. Local development involves npm install @gannonh/memento-mcp or cloning the repository.
• Prerequisites: Neo4j 5.13+ (required for vector search), Node.js/npm, and an OpenAI API key for semantic search capabilities. Neo4j can be set up via Neo4j Desktop or Docker Compose.
• Setup: Involves installing Neo4j, configuring environment variables (including `OPENAI_API_KEY