mcp-zeromem
Conversational memory for agents that costs zero tokens to maintain.
Every turn an agent sees goes into one SQLite store. Recall over that store — an entity graph plus lexical and dense search — runs with no LLM in the loop, so remembering is free: the only tokens spent are the ones the model spends reading what came back. That is the whole idea, and it is what separates this from memory systems that pay a summarization call on every write.
The method follows the Zero-Mem paper (arXiv:2607.29377), whose upstream implementation is ptaranat/zeromem.
Stack: TypeScript, SQLite, MCP.
Full documentation lives in the project README: https://github.com/cubicecho/mcp-zeromem