
MCP Tasks: Long-Running Tool Calls Without Timeouts
Summary
Build an async MCP tool that returns a durable task handle and poll it to completion in Python.
On July 28, 2026 the Model Context Protocol shipped the release candidate that developers had been waiting on all summer, and the headline feature is Tasks: a standard way for an MCP tool to run for minutes or hours without holding a connection open. Until now, an agent that called a tool waited, blocked, for the answer. If the work took longer than a transport timeout, the call died. Tasks fix that by letting the server hand back a durable handle instead of the result, and letting the client poll for progress and pick up the answer later, even after a reconnect.
This matters right now because agents are moving off the demo stage and into real infrastructure: CI pipelines, batch jobs, deployments, and human approval gates. Every one of those is longer than the few seconds a blocking HTTP call can survive. Tasks are the protocol-level primitive that makes those workflows first-class instead of hacks bolted on with side channels.
Keep reading — it's free
Enter your email to keep reading — plus the best of AI & tech, daily. Free, forever.
Already a member? Sign in
Comments
Be the first to comment