Testing Long-Running Remote MCP Tasks: What the Agent Actually Receives

On October 3, 2026, DEV Community published a developer write-up on a question that MCP demos rarely address: what an agent actually receives when a remote MCP tool runs for minutes instead of answering within a second.

The author points out that most MCP demos show a tool returning an answer almost instantly, while real tools often behave differently. A scraper, a crawl, or a report job can take minutes, and the team builds web scrapers on Apify, so this is a problem they run into daily.

To find out what happens on the wire, the developer traced the raw JSON-RPC traffic against Apify's hosted MCP server at mcp.apify.com, which was on version 0.17.1 at the time. The test task was a YellowPages scraper that reads only a single page of results, with an overall run time close to that of a real scraper, a batch crawl, or a report job.

The write-up stops at the test setup. It names the traced server and its version, the task, and the fact that the run lasted minutes rather than one call, but it does not report the specific traffic contents, does not draw conclusions, and does not describe any follow-up fix.