@aiagentkarl/agent-memory-mcp-server

Persistent memory for AI agents. Store, retrieve, and search knowledge across sessions using the Model Context Protocol.

Features

• Persistent Storage -- Key-value pairs stored as JSON files in ~/.agent-memory/
• Namespaces -- Organize memories into separate namespaces (e.g. project_x, user_prefs)
• Metadata -- Attach arbitrary metadata to each memory entry
• Search -- Substring search across keys, values, and metadata
• Access Tracking -- Automatic access counts and timestamps

Tools

| Tool | Description | |------|-------------| | store | Store a key-value pair with optional metadata | | retrieve | Retrieve a stored value by key | | search | Search across memories using substring matching | | list_keys | List all keys in a namespace | | delete | Delete a stored entry | | list_namespaces | List all available namespaces |

Installation

npm install -g @aiagentkarl/agent-memory-mcp-server

Usage with Claude Desktop

Add to your claude_desktop_config.json:

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

Usage with Claude Code

claude mcp add agent-memory -- npx -y @aiagentkarl/agent-memory-mcp-server

Storage

Data is stored as JSON files in ~/.agent-memory/. Each namespace gets its own file:

~/.agent-memory/
  default.json
  project_x.json
  user_prefs.json

Set MEMORY_STORAGE_DIR environment variable to change the storage location.

License

MIT