memory

Vector-based memory storage with embeddings for semantic similarity search.

When to use memory

• Store important context for future sessions
• Recall relevant information from past conversations
• Build persistent knowledge about user preferences
• Semantic search across stored memories

Available MCP tools

| Tool | Purpose | |------|---------| | mcp__memory__store | Store a memory | | mcp__memory__search | Semantic similarity search | | mcp__memory__list | List recent memories | | mcp__memory__delete | Remove a memory | | mcp__memory__get_context | Get relevant context |

Common patterns

Store a memory

mcp__memory__store(
  content="User prefers TypeScript over JavaScript for new projects",
  tags=["preferences", "programming"]
)

Search memories

mcp__memory__search(query="What does the user prefer for web development?", limit=5)

Get context for a topic

mcp__memory__get_context(topic="user's coding preferences")

List recent memories

mcp__memory__list(limit=10)

Best practices

1. Store preferences - When user expresses a preference, store it
2. Store decisions - Important decisions and their rationale
3. Store context - Project-specific context that spans sessions
4. Search before assuming - Check memory before making assumptions

CLI commands (if MCP unavailable)

memory add "Content to remember" --tag preference
memory search "query"             # Semantic search
memory list                       # Recent memories
memory export -f markdown         # Export

Data location

~/.local/share/memory/memory.db (SQLite with embeddings as BLOB, respects XDG_DATA_HOME)