Memory

The Memory MCP Server is part of the official MCP server collection maintained by Anthropic. It provides persistent memory across conversations using a local knowledge graph — giving your AI assistant the ability to remember facts, people, preferences, and relationships between sessions.

How it works

The server stores a knowledge graph in a local JSONL file with three primitives:

Entities — Named nodes with a type and observations:

jsonCOPY
{
  "name": "John_Smith",
  "entityType": "person",
  "observations": ["Speaks fluent Spanish", "Prefers dark mode"]
}

Relations — Directed connections between entities:

jsonCOPY
{
  "from": "John_Smith",
  "to": "Acme_Corp",
  "relationType": "works_at"
}

Observations — Atomic facts attached to entities. One fact per observation keeps things clean and deletable.

Custom storage location

By default, the knowledge graph is stored as memory.jsonl in the server's working directory. To use a specific path:

jsonCOPY
{
  "mcpServers": {
    "memory": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-memory"],
      "env": {
        "MEMORY_FILE_PATH": "/Users/you/.ai-memory/memory.jsonl"
      }
    }
  }
}

This is useful when you want multiple AI tools to share the same memory file, or when you want to back up your knowledge graph with your dotfiles.

Prompting for better memory

The server works best when the AI is instructed to actively use it. Add instructions to your system prompt or Claude Project:

Follow these steps for each interaction:

1. Always begin by retrieving relevant information from your memory
2. While conversing, note new information about:
   - People and their preferences
   - Project requirements and decisions
   - Technical choices and their rationale
3. After each interaction, update memory with any new facts learned

Docker installation

jsonCOPY
{
  "mcpServers": {
    "memory": {
      "command": "docker",
      "args": ["run", "-i", "-v", "claude-memory:/app/dist", "--rm", "mcp/memory"]
    }
  }
}

The Docker volume claude-memory persists the knowledge graph between container restarts.

Common issues

Memory resets between sessions Check that MEMORY_FILE_PATH points to a persistent location. If you're using npx without setting this, the file may be created in a temporary directory that gets cleaned up.

"Entity not found" errors Entity names are case-sensitive. John_Smith and john_smith are different entities. Use consistent naming conventions.

Graph getting too large The server loads the entire graph into memory. For very large knowledge graphs (10,000+ entities), searches may slow down. Periodically prune outdated entities.