modelcontextprotocol/server-memory
Free官方参考实现，基于知识图谱的持久记忆系统。
FreeFree tier
About modelcontextprotocol/server-memory
An official reference implementation of persistent memory using a local knowledge graph, published as @modelcontextprotocol/server-memory. It enables Claude to remember information about users across chats by storing entities, relations, and observations in a structured, searchable graph. Entities have types and observations, relations are directed connections in active voice. Provides tools for creating, reading, updating, and deleting entities, relations, and observations, as well as searching the graph. Designed for integration with the Model Context Protocol (MCP).
Key Features
Knowledge graph memory system with entities, relations, and observations
Local persistence for cross-chat recall
Tools to create, read, update, delete entities and observations
Relation management with active voice directed connections
Search across entity names, types, and observation content
Published as npm package @modelcontextprotocol/server-memory
Compatible with Model Context Protocol (MCP) servers
Pros & Cons
Pros
- Open source and free to use
- Structured knowledge graph with clear entity, relation, observation model
- Easy integration with MCP-compatible clients like Claude
- Provides full CRUD operations plus search
- Supports atomic observations for granular memory
- Well-documented API and example usage
Cons
- Local storage only - no built-in distributed or cloud persistence
- Limited to the memory features provided; not a general-purpose database
- Requires running a local MCP server instance
- No built-in authentication or multi-user support
Best For
Enabling AI assistants to remember user preferences and history across conversationsStoring structured knowledge about users (e.g., names, affiliations, preferences)Building persistent context for chatbots and virtual assistantsResearch and experimentation with graph-based memory in MCP
FAQ
What is the Knowledge Graph Memory Server?
It's an official MCP reference implementation that provides persistent memory using a local knowledge graph, allowing Claude to remember information across chats.
How do entities, relations, and observations work?
Entities are nodes with a name, type, and list of observations (strings). Relations are directed connections between entities described in active voice. Observations are atomic facts attached to entities.
What tools are available?
Tools include create_entities, create_relations, add_observations, delete_entities, delete_observations, delete_relations, read_graph, and search_nodes.
How can I install and use this server?
It's published as @modelcontextprotocol/server-memory on npm. You can install it and configure it as an MCP server in your client application.