We've Been Through This -- scalable smart memory mcp
WeaveHacks 3: Self-Improving Agents Hackathon with Weights & BiasesProblem We lose hours and millions of tokens getting stuck on conceptually similar issues in Claude Code. It gets even worse in teams where knowledge is shared ad hoc. Solution A team-shared MCP that conceptualizes where you get stuck and remembers. A manual workaround is to add lessons to AGENTS.md, create skills, or write custom prompts - but a self-learning system is the natural evolution. To do it properly, you need to study history, extract knowledge, store it for the whole team, and inject it into the prompt. That is what we do: trace Claude with Weave, analyze, store in Redis, and inject back into Claude. The memory is hybrid-smart: the AI figures out which schemas and templates to use.
