Memento — a local-first MCP server that gives your AI durable repository memory

I’ve been experimenting a lot with local AI coding workflows, and I kept running into the same problem: Even with large context models, repositories are still far bigger than the context window. After a few prompts the model forgets: architecture decisions relationships between modules previous exploration of the codebase design notes or reasoning So you end up re-explaining the same things over and over. I built Memento to try to solve that. What it is Memento is a local-first MCP server that gives AI agents durable memory about a repository.