task-patterns

To manage long-running MCP tool executions without blocking, you return Task IDs instead of synchronous responses. Clients then use tasks/get, tasks/list, and tasks/result to poll status and retrieve deferred outcomes asynchronously.

The best way to cancel an asynchronous MCP task is by applying task_support patterns to your tools, which enforce a TaskStorage-backed state machine. This enables a cancelable workflow across the task lifecycle using your chosen storage backend.

Yes, MCP task management works with persistent storage backends like PostgreSQL, DynamoDB, and SQLite. You integrate these with the TaskRuntime and TaskExecutor to enforce reliable state machine lifecycle management.

To configure an MCP server for asynchronous task tracking, you set task_support patterns (optional, required, or forbidden) on your tools. Starting a task-enabled server then applies the TaskStorage-backed state machine for lifecycle management.

You should use asynchronous task IDs instead of direct synchronous responses for long-running tool operations. Returning Task IDs enables status updates, cancellation, and deferred results, which are complex to manage synchronously without blocking the client.

Yes, you can enforce a specific task lifecycle state machine across different storage backends in MCP. The skill applies task_support patterns to integrate TaskRuntime with InMemoryTaskStorage, SQLite, PostgreSQL, or DynamoDB for consistent state transitions.