Enhance Claude with Persistent Memory

The Memory MCP server provides Claude with persistent memory using a local knowledge graph. This lets Claude remember information about users across chats through entities, relations, and observations.

Installation

npm install -g @modelcontextprotocol/server-memory

Configuration

Add to your Claude Code settings:

{
  "mcpServers": {
    "memory": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-memory"
      ]
    }
  }
}

Custom Memory File Path

{
  "mcpServers": {
    "memory": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-memory"],
      "env": {
        "MEMORY_FILE_PATH": "/path/to/custom/memory.jsonl"
      }
    }
  }
}

Core Concepts

Entities

Primary nodes in the knowledge graph with a unique name, type, and observations.

{
  "name": "John_Smith",
  "entityType": "person",
  "observations": ["Speaks fluent Spanish"]
}

Relations

Directed connections between entities in active voice.

{
  "from": "John_Smith",
  "to": "Anthropic",
  "relationType": "works_at"
}

Observations

Discrete, atomic pieces of information attached to entities.

Available Tools

create_entities

Create multiple new entities in the knowledge graph.

create_entities(entities: Entity[]): void

create_relations

Create relations between entities.

create_relations(relations: Relation[]): void

add_observations

Add new observations to existing entities.

add_observations(observations: ObservationInput[]): AddedObservations[]

delete_entities

Remove entities and their associated relations.

delete_entities(entityNames: string[]): void

delete_observations

Remove specific observations from entities.

delete_observations(deletions: DeletionInput[]): void

delete_relations

Remove specific relations from the graph.

delete_relations(relations: Relation[]): void

read_graph

Read the entire knowledge graph.

read_graph(): KnowledgeGraph

search_nodes

Search for nodes by query across names, types, and observations.

search_nodes(query: string): SearchResults

open_nodes

Retrieve specific nodes by name with their relations.

open_nodes(names: string[]): NodeResults

Recommended System Prompt

Add this to your Claude project for optimal memory usage:

Follow these steps for each interaction:

1. User Identification:
   - Assume you are interacting with default_user
   - Proactively try to identify the user if unknown

2. Memory Retrieval:
   - Begin chats by saying "Remembering..." and retrieve relevant information
   - Refer to the knowledge graph as your "memory"

3. Memory Updates:
   - Be attentive to new information: identity, behaviors, preferences, goals, relationships
   - Create entities for recurring people, organizations, and events
   - Connect them using relations and store facts as observations

Usage Example

Claude, remember that I prefer morning meetings and
my favorite programming language is Python.