Buidling ZINQ AI , customer support AI agent! with MCP support

Interesting pivot from information collection into a full multi-agent support system. A lot of customer support AI products still struggle once conversations span channels or require coordination between retrieval, action execution, and escalation logic, so the MCP support angle caught my attention. Curious whether your agents share a unified memory/context layer across channels or if each interaction is handled independently.

Also interested in how you’re thinking about human handoff and failure recovery — especially for voice workflows where latency and incorrect actions become much more noticeable than in chat.