Memento MCP is a scalable, high-performance knowledge graph memory system designed for LLMs. It provides long-term ontological memory with semantic retrieval, contextual recall, and temporal awareness, enabling any MCP client that supports the model context protocol to access persistent memory. Built on Neo4j, it unifies graph storage and vector search, delivering efficient semantic search through vector embeddings (OpenAI embedding models) and cosine similarity, with configurable thresholds and cross-modal search. Entities are core nodes with embeddings and complete version history; relations connect entities with strength, confidence, and rich metadata, and are versioned over time. Features include non-destructive updates, time-based filtering, and change tracking. The MCP API exposes tools for managing entities, observations, relations, and graph operations (read_graph, search_nodes, open_nodes) along with semantic search and embedding retrieval. Integration with Claude Desktop is documented, with example configurations for both real and local development. Advanced capabilities include confidence decay, adaptive hybrid search strategies, and extensive metadata support, plus diagnostics and debugging tools for development and operations. The project emphasizes persistence, resilience, and semantic memory for memory-enabled LLM workflows.