Memento MCP

Memento MCP is an MCP server that provides large language models with persistent, long-term ontological memory by interfacing with a Neo4j graph database. Software engineers, AI researchers, and agent developers use it to equip compatible MCP hosts, such as Claude Desktop, Cursor, and GitHub Copilot, with structured knowledge graph storage and vector-based semantic retrieval. The server organizes context into entities and directed relationships, combining property graph architecture with OpenAI vector embeddings for hybrid retrieval. Memento MCP supports temporal awareness, preserving a complete version history of entity modifications and timestamped relationships, which allows point-in-time state exploration. In addition to historical tracking, it incorporates time-based confidence decay with configurable half-lives, simulating natural forgetting when data is not reinforced over time. By unifying graph traversal, metadata tracking, and embedding search in Neo4j 5.13+, the server enables LLM agents to maintain cross-session context, track evolving real-world facts, and update relational data through dedicated tool calls such as creating entities, managing observations, and traversing connected graphs.

Category: AI Memory & Context

Tags: knowledge-graph, memory, neo4j, rag, semantic search

Visit Memento MCP

How to install and configure Memento MCP

1. Deploy a Neo4j database instance (version 5.13 or higher is required for vector search). You can launch it with Docker using docker-compose up -d neo4j or configure a database in Neo4j Desktop accessible at bolt://127.0.0.1:7687 with default password memento_password. 2. Test connectivity to the database from the project directory by executing npm run neo4j:test. Schema initialization occurs automatically on first connection, or you can run npm run neo4j:init manually. 3. Configure your MCP client (such as Claude Desktop or Cursor) by following the client registration guidelines and environment configurations detailed in the repository at https://github.com/gannonh/memento-mcp.

What you can do with Memento MCP

• Building persistent developer agent memory to track code architecture decisions, team roles, and dependency changes across multiple chat sessions. - Storing evolving real-world knowledge graphs where past states can be inspected using timestamped point-in-time graph queries. - Modeling temporal relationships with automated confidence decay to reflect outdated facts when information is not reinforced periodically. - Performing hybrid semantic and keyword search across unstructured agent observations using integrated Neo4j vector embeddings. - Maintaining cross-session entity profiles and relationship networks for AI assistants operating inside Claude Desktop or Cursor.

Key facts

• Open Source
• https://github.com/gannonh/memento-mcp
• AI Memory & Context, Databases & Data Stores
• knowledge-graph, memory, neo4j, rag, semantic search

Part of MCP Servers

What is Memento MCP?

Memento MCP is an open source Model Context Protocol server that equips LLM clients with long-term memory via a Neo4j graph database. It manages entities, observations, and directed relationships, combining structured graph data with vector embeddings for semantic search and point-in-time historical recall.

Which MCP clients work with Memento MCP?

Memento MCP works with any AI client host that supports the Model Context Protocol. Supported applications explicitly mentioned by the project include Claude Desktop, Cursor, and GitHub Copilot, allowing these tools to read and write graph-based memories persistently.

What database is required to run Memento MCP?

Memento MCP requires Neo4j version 5.13 or later. This version is necessary because Memento MCP relies on Neo4j for unified storage, executing native property graph operations alongside integrated vector similarity search for OpenAI text embeddings.

How does temporal awareness work in Memento MCP?

Temporal awareness in Memento MCP preserves a complete version history whenever entities or relations are updated. It stores creation, update, and validity timestamps, enabling point-in-time queries to inspect how the graph looked in the past without destroying older historical context.

Is Memento MCP open source?

Yes, Memento MCP is an open-source project hosted on GitHub. Developers can inspect its source code, contribute to its development, test connections with provided scripts, or run it self-hosted using Docker and Neo4j.