Memory MCP Server

MCP Server Configuration

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

Optionally specify a custom storage path:

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

Available Tools

Tool	Description
create_entities	Add new entities to the knowledge graph
create_relations	Create relationships between entities
add_observations	Add observations/facts to existing entities
delete_entities	Remove entities and their relationships
delete_observations	Remove specific observations from entities
delete_relations	Remove relationships between entities
read_graph	Read the entire knowledge graph
search_nodes	Search for entities by name or content
open_nodes	Retrieve specific entities by name

Example Usage

# Store project architecture knowledge
create_entities: [
  { name: "AuthService", entityType: "service", observations: ["Handles JWT tokens", "Uses bcrypt for passwords"] },
  { name: "UserRepository", entityType: "repository", observations: ["PostgreSQL backed", "Has caching layer"] }
]

# Create relationships
create_relations: [
  { from: "AuthService", to: "UserRepository", relationType: "depends_on" }
]

# Later, search for context
search_nodes: { query: "authentication" }

Setup

1. Add the configuration to .claude/settings.json
2. No API keys required
3. Memory persists to ~/.claude/memory.json by default
4. Set MEMORY_FILE_PATH for project-specific memory

Use Cases

• Project knowledge — Store architecture decisions, conventions, gotchas
• User preferences — Remember coding style, preferred libraries, workflow
• Entity relationships — Map service dependencies, data flows, team ownership
• Session continuity — Pick up context from previous sessions automatically

Security Notes

• Data stored locally in a JSON file — not sent to any external service
• Memory file may contain sensitive project details — treat it like code
• Add the memory file to .gitignore if it contains private information
• No network access — purely local storage
• File is human-readable JSON — can be manually inspected and edited