Memory Mcp

Adds persistent memory to Claude Code by automatically saving and retrieving conversation context across sessions. Uses vector embeddings to store two types of knowledge: episodes (full conversation sessions with summaries) and entities (extracted decisions, preferences, and concepts). Integrates via hooks in Claude Code's settings.json to auto-save messages on UserPromptSubmit and Stop events, then injects relevant history from past sessions as context. Exposes tools like memory_recall for semantic search, memory_add_entity for manual knowledge capture, and memory_start_episode for session management. Storage splits between user-level (~/.claude-memory) for cross-project preferences and project-level (.claude/memory) for project-specific decisions. Requires a one-time 400MB model download and Python 3.10-3.13. Useful when you want Claude to remember architectural decisions, coding preferences, or project context without re-explaining everything each session.