mai-mcp
Memory that outlives the session
A multi-project memory server for AI agents. Decisions, lessons and code structure persist across sessions, models and tools — so the context does not die when the window fills.
Every AI coding session starts from zero. Context windows fill, sessions end, models change — and everything the agent worked out last time is gone.
mai-mcp is the riverbed under that flow. It keeps a project's decisions, lessons and code structure in a database that any agent can read and write through the Model Context Protocol, so the reasoning behind a choice survives long after the conversation that produced it. Writes are gated behind structured evidence, so the record stays trustworthy rather than filling with noise.
Each server instance is pinned to a single project — agents cannot reach across project boundaries. It runs against Anthropic or OpenAI models, or entirely locally through Ollama, and ships with a command-line interface and a review dashboard.