Daem0n-MCP is a Model Context Protocol server that gives AI agents durable memory across sessions. Instead of bloated markdown files that eat into your context window, it stores and retrieves knowledge through an active memory layer designed for efficiency. Token compression cuts memory overhead by up to 75%, keeping costs down on long-running agents. Passive capture hooks record decisions automatically without requiring explicit save commands. Causality tracking links outcomes to past choices, and a built-in failure-weighting system boosts the salience of failed decisions by 1.5x — so the agent is more likely to reconsider a bad pattern before repeating it. Useful for developers building autonomous agents, personal AI assistants, or any MCP-compatible workflow where continuity and learning across sessions matter.