memory_map v1.0.0 — Persistent Memory & History for AI

A persistent memory and conversation history server for Claude Code, built on the Model Context Protocol (MCP). It gives Claude a long-term memory across sessions, no more re-explaining your stack, preferences, or in-progress work at the start of every conversation.

Features

• 18 MCP tools exposed via FastMCP: save, load, delete, query, and compress per-project memory
• Conversation history stored in MongoDB with semantic search (OpenAI or Atlas autoEmbed) and keyword tag-matching fallback
• Global memory for user-level facts shared across all projects
• Cross-project scanning: list, search, and compare memory across multiple repos
• Local file tree + GitHub structure tools for rapid codebase orientation
• Automatic history hook: fires on every prompt and pre-compaction to summarise and persist conversations
• Atomic writes with portalocker throughout — safe under concurrent access
• Compression levels (raw / compact / dense) to control context budget

Requirements

• Python 3.11+
• MongoDB Atlas (for conversation history and vector search)
• OpenAI API key (optional for embedding-based semantic search)
• Claude Code CLI

Setup

See README.md for full installation and configuration instructions