Category-based Memory Server

What is Category-based Memory Server?

The Category-based Memory Server provides persistent memory for Claude using a category and tag system. It stores memories as plain text files, supporting global storage (shared across all projects) and local storage (specific to the current project).

How to use Category-based Memory Server?

Install via npx and configure in claude_desktop_config.json or VS Code settings. Use the provided tools (remember_memory, retrieve_memories, remove_memory_category, remove_specific_memory) to store, retrieve, and manage categorized memories. The MEMORY_FILE_PATH environment variable can set a custom storage location.

Key features of Category-based Memory Server

• Organize memories by categories and tags.
• Supports global (home directory) and local (project) storage.
• Memories stored as plain text files.
• Add and remove memories independently.
• Atomic memories (one piece of information per entry).

Use cases of Category-based Memory Server

• Remember coding preferences and conventions across chat sessions.
• Store project‑specific knowledge that persists across interactions.
• Maintain a shared knowledge base for consistent behavior.

FAQ from Category-based Memory Server

What is the difference between global and local memories?

Global memories are stored in the user’s home directory (~/.config/goose/memory) and persist across all projects. Local memories are stored in the current project directory (.goose/memory) and are specific to that project.

What file format are memories stored in?

Memories are stored as plain text files. Each category has its own file, and tags are prefixed with # (e.g., # formatting style).

How do I configure a custom memory file path?

Set the MEMORY_FILE_PATH environment variable when starting the server (default is memory.json in the server directory).

Can I remove all memories at once?

Yes, use the remove_memory_category tool with * as the category, and specify whether to target global or local storage.

What are the runtime requirements?

The server requires Node.js and is distributed via npm as @mkusaka/mcp-server-memory. It communicates via MCP stdio and does not require authentication.