Knowledge Graph Memory MCP Server

by Anthropic (MCP reference servers)

Reference Agent Reasoning & Memory

A reference server that gives an agent persistent memory across sessions, stored as a simple local knowledge graph of entities, relations, and observations.

Add to Claude Code

claude mcp add memory -- npx -y @modelcontextprotocol/server-memory

Source: https://github.com/modelcontextprotocol/servers/tree/main/src/memory

About

The Memory server persists what an agent learns about entities (people, projects, preferences) and the relationships between them as a lightweight local knowledge graph, stored in a JSON file — so context survives between separate conversations instead of resetting every session.

It's intentionally simple: entities have observations attached to them, and relations connect entities to each other, which is enough structure for an agent to build up and query a working memory of ongoing context.

Key features

  • Persistent memory across separate agent sessions
  • Simple entity-relation-observation knowledge graph model
  • Stored locally in a plain JSON file — fully inspectable
  • Searchable by entity name or observation content

Use cases

  • Let an agent remember a user's preferences and project context between sessions
  • Build up a running knowledge base of facts learned during a long research task
  • Track relationships between people, projects, and decisions over time

Available tools

create_entities

Creates new entities with a name, type, and initial observations.

create_relations

Creates a directional relationship between two existing entities.

search_nodes

Searches entities and observations by keyword.

Frequently asked questions

Where is the memory stored?

In a local JSON file on disk — fully readable and editable outside the agent if needed, with no external database required.

Does it sync across devices?

Not on its own — it's a local file, so syncing would require your own file-sharing or backup setup.