MCP Async Tasks: Building long-running workflows for AI Agents

The 2025 revision of the Model Context Protocol (MCP) introduces "Tasks," an experimental feature that transforms MCP from a synchronous model into a call-now, fetch-later protocol, allowing asynchronous operations. Tasks address common issues like timeouts and blocked sessions by standardizing long-running operations, thus enabling seamless interoperability among clients, servers, and SDKs. This upgrade allows for better parallelization, consistent progress reporting, and efficient handling of extensive workloads, such as ETL jobs and large file conversions, by introducing a cross-request asynchronous state machine. Although tasks are currently experimental, they are already being actively implemented across languages, as they are foundational for managing complex workflows and enhancing user experience. Security considerations are emphasized, as tasks are fetched later by task IDs, requiring robust authorization mechanisms to ensure safe and compliant execution. Overall, Tasks are set to become a fundamental aspect of MCP, offering a more powerful and flexible approach to managing asynchronous operations within the protocol.