Dakera Mcp

Connects Claude or any MCP client to a Dakera memory instance for persistent agent memory with decay-weighted vector storage. Ships with 14 high-frequency tools by default (store, recall, search, sessions, knowledge graphs) to keep token overhead around 3K, then lets you discover and load 72 more on demand from admin, power, and all tiers. The tiered profile system means you pay token cost only for the operations you actually need. Requires a running Dakera server, either self-hosted via Docker or their managed instance. Scores 88.2% on the LoCoMo benchmark. Useful when you need agents to remember context across sessions without hardcoding retrieval logic or burning 15K tokens on unused tool schemas.