Memento MCP

Memento MCP is a high‑performance knowledge‑graph memory system built to give language‑model clients persistent, semantic recall. By exposing a Model Context Protocol (MCP) interface, it lets assistants such as Claude Desktop, Cursor, or GitHub Copilot maintain a durable ontological memory that can grow with every conversation. The core problem it solves is the short‑term nature of most LLMs: without a separate, structured memory layer, assistants lose context after each session or rely on fragile prompt engineering. Memento MCP stores entities, relations, and their observations in a Neo4j graph, enabling instant retrieval of facts, relationships, and temporal updates even across large datasets.

At its heart the server manages entities—named nodes with types, observations, embeddings, and full version histories. When a new piece of information is added, it becomes an observation attached to the relevant entity, and a vector embedding is generated for semantic search. Relations connect these entities with rich metadata: strength, confidence, source tags, and timestamps. The system also models time‑based decay in confidence, so older facts can be downgraded automatically unless refreshed. These features give developers a powerful way to encode complex, evolving knowledge without writing custom persistence logic.

Key capabilities include:

• Semantic retrieval through vector search integrated into Neo4j, allowing fuzzy matches on natural‑language queries.
• Temporal awareness that tracks version history and applies confidence decay, ensuring the assistant reflects the most current state of knowledge.
• Unified storage where graph traversal and vector similarity are handled by the same database, reducing latency and simplifying architecture.
• Scalable performance thanks to Neo4j 5.13+ vector support, enabling millions of entities and relations without compromising query speed.

Typical use cases span personal knowledge bases (tracking contacts, projects, or learning progress), enterprise data integration (linking employees to products and events), and context‑aware recommendation systems where the assistant must remember user preferences over time. In a workflow, a developer writes MCP‑compatible commands to insert or query entities; the assistant then enriches user interactions with up‑to‑date, semantically relevant information pulled directly from the graph.

What sets Memento MCP apart is its blend of structured ontology and semantic search, all wrapped in a protocol‑friendly interface. Developers can quickly bootstrap persistent memory for their AI applications, avoiding the pitfalls of prompt‑based state management while keeping the system lightweight and extensible.