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

A long-running AI tool needs a way to resume after the connection that started it disappears. Temporal's presentation on MCP Tasks explains this through invoice processing that pauses for human approval. MCP, a protocol connecting agents to tools, can return a task identifier immediately while the operation continues in the background. The client uses that identifier to follow the task and obtain its eventual result. Making this dependable requires stored task state on the service side and enough client bookkeeping to reconnect to the right work. Simply returning an identifier does not supply that persistence; the application must preserve it across process failures and interruptions.

“stateful protocols are the absolute worst thing in large-scale distributed systems”

Cornelia Davis

“Spoiler alert, there is no filter on that endpoint. So, you would have to go through a million tasks to find the one that you're looking for”

Cornelia Davis

“once you've locked launched a task it has to be durable”

Cornelia Davis

“if I've got a million tasks running, I've got a million clients that are doing gets against each and every one of those tasks. That does not scale.”

Cornelia Davis

“Just because you can doesn't mean you should.”

Cornelia Davis