AI
Anthropic adds durable background jobs and human prompts to Claude Code
Anthropic has added MCP Tasks and Elicitations to Claude Code, allowing compatible servers to run jobs beyond a normal tool-call timeout and request human input while work is underway. Tasks return durable handles that can survive a client exit and reconnect, while Elicitations support confirmations, forms and browser-based authentication. The capabilities are available in the CLI, Agent SDK and GitHub Actions, but public-beta users must handle persistence, reconnect ambiguity, transport changes and cancellation carefully.
Editorial summary · Publisher article
What matters
- Tasks let servers return a handle quickly and report progress through polling rather than keeping the original connection open.
- Elicitations can pause a task for approval or missing information, while secrets should use URL-based authentication flows.
- Servers must persist task state and make retries, completion and cancellation idempotent.
- One-shot runs cannot create MCP Tasks, and non-interactive elicitations cancel the task.
Context & caveats
- The account is based on Anthropic’s implementation and documentation; broader Claude-surface availability is described as forthcoming.
Why this made the edition
Detailed technical reporting explains how MCP Tasks and Elicitations change long-running AI tool workflows, including compatibility requirements and operational failure modes.
Source material