Memory Management

Core principle: Store valuable learnings. Recall relevant context. Build institutional knowledge across sessions.

MCP Tools

memory_store

Store a memory for future recall.

memory_store(
  content: string,       # What to remember
  type?: "fact" | "entity" | "relationship" | "self",
  importance?: 0-10,     # 8+ = critical, 5-7 = high, 3-4 = normal
  tags?: string[],       # For categorization
  context?: string       # Additional context
)

When to store:

• •User preferences ("I prefer tabs over spaces")
• •Project decisions ("We chose PostgreSQL for X reason")
• •Architecture patterns ("Auth flow uses JWT with refresh tokens")
• •Lessons learned ("Don't use library X, it has issue Y")
• •Entity relationships ("User manages Project, Project has Tasks")

memory_recall

Search and retrieve relevant memories.

memory_recall(
  query: string,         # What to search for
  type?: string,         # Filter by memory type
  limit?: number,        # Max results (default 10)
  token_budget?: number  # Max tokens to return
)

When to recall:

• •Starting work on a project (recall project context)
• •Making decisions (recall past decisions and rationale)
• •User asks "remember when..." or "what did we decide about..."
• •Before suggesting solutions (recall past learnings)

memory_forget

Remove a memory (soft delete with provenance).

memory_forget(id: string)

Best Practices

1. •Be selective - Store valuable, reusable knowledge, not ephemeral details
2. •Use appropriate importance - 8+ for critical decisions, 5-7 for useful info
3. •Add context - Include WHY something matters, not just WHAT
4. •Recall proactively - Check for relevant context before making suggestions
5. •Update over time - Store corrections when past memories become outdated

Memory Types