Apify uses Standby mode to keep MCP servers always-on without polling

On November 7, 2026, DEV Community's AI section published an article that introduces Apify's google-news-mcp-server and walks through how it works differently from the rest of the author's Apify Actors. According to the author, every other Actor built this way follows the same shape: start a run, poll GET /v2/actor-runs/<id> until the status reaches SUCCEEDED, then fetch the dataset. That pattern is fine for batch work, the author notes, since when a caller asks for 5,000 rows nobody minds a startup cost of roughly 10 seconds. The problem is the agent that asks a single live question mid-conversation: with the start-a-run-and-poll shape, that latency falls on the agent's user, who has to sit through it in real time.

google-news-mcp-server takes a different approach. It runs in Apify Standby mode, which turns the Actor into a long-lived Web server bound to a fixed URL. A client sends MCP tools/call requests over Streamable HTTP and receives the results back on that same connection, so there is no need to start a run first and poll for completion. The author reports testing tools/list on a cold server and measuring 3.99 seconds until an HTTP 200 response was returned. No earlier announcements or later developments appear in the available material, so the account currently ends at this description of the mechanism and the cold-start measurement.