Agent Loop Mcp

This is a persistent state manager for long-running agent workflows that you install in two parts: an MCP server for storage and a skill file that teaches the agent when to compact memory. It monitors word counts to trigger compression cycles and includes a self-healing strategy to break infinite loops when agents get stuck. You'd reach for this when working with smaller models that have limited context windows but need to maintain state across extended tasks. The server exposes standard memory operations through stdio transport, while the skill instructions live in SKILL.md and orchestrate when to save, retrieve, and compress context. Think of it as giving any model the persistent memory behavior of larger ones without requiring massive context.