MCP Tasks (async): Why Aren't Any Agents Supporting Them? — Cornelia Davis, Temporal

MCP Tasks represent a critical evolution in agentic workflows by enabling long-running, asynchronous processes that survive infrastructure failures and human-in-the-loop delays. The current lack of widespread client support stems from the high complexity of implementing durable state management, as demonstrated by the challenges of handling network blips, process crashes, and session persistence. Version 1 of the specification relies on stateful protocols and cumbersome task-listing mechanisms that struggle to scale. The upcoming Version 2 addresses these issues by shifting to a stateless core, removing the inefficient task-list endpoint, and introducing a more robust signaling mechanism for interaction. Successful implementation requires developers to map task lifecycle states to domain-specific state machines, ensuring that even when infrastructure disconnects, processes like purchase order management can resume seamlessly without data loss or manual intervention.