Exposes four MCP tools to monitor and protect Markdown vaults used as agent memory. Predict_rot scores your corpus on five context degradation factors before you inject it into an LLM. Get_tiered_read splits notes into working, episodic, and semantic tiers based on recency. Record_journal and read_journal maintain an Ed25519-signed, hash-chained audit log of every agent write, which you verify with the CLI to catch tampering. Ships with a Claude Code PostToolUse hook that passively signs writes as they happen. Useful when you're growing a vault with an agent over months and need to know whether the corpus is rotting, what subset to load, and whether anyone has modified the history.
claude mcp add --transport stdio toonight-mnemoscope uvx mnemoscope