Long-running jobs and mid-task questions reach Claude Code
Anthropic said on October 23 that Claude now supports two parts of the Model Context Protocol, Tasks and Elicitations. Both work today in the Claude Code CLI, and the company says they will reach more Claude products soon.
Tasks let Claude hand a plugin a long-running job, with deep research as Anthropic’s example, and the plugin reports back once the job is done. Elicitations go the other way: in the middle of a task, a plugin can stop and ask the user for something, such as confirming a refund or filling in a missing detail. Anthropic’s MCP documentation for Claude Code now has sections for server authors on adding Tasks and on responding to elicitation requests.
Neither feature is Anthropic’s invention. The 2025-11-25 revision of the MCP specification added tasks as an experimental feature for tracking durable requests: the client polls, and the result can be fetched later. The same revision expanded elicitation with a URL mode, enum schemas that can be titled or untitled and single- or multi-select, and default values for strings, numbers and enums.
The protocol has changed again since then. The 2026-07-28 specification took tasks out of the core protocol and made them an official extension, io.modelcontextprotocol/tasks. In that extension, clients poll with tasks/get where the blocking tasks/result method used to be, a new tasks/update carries input from the client to the server, tasks/list is gone, and a server can return a task handle without the client opting in on each request. Elicitation also works differently in the new version. The changelog in the spec repository says servers no longer send elicitation/create as a request of their own. A server answers the original request with a result whose resultType is "input_required" and lists what it needs. The client then retries that request with the answers attached.
The announcement is aimed at plugin builders. In Anthropic’s setup, a plugin is the package that bundles MCP servers and skills, and Anthropic calls it the main way third-party developers extend Claude. Its documentation describes the plumbing such a server plugs into. Remote servers connect over HTTP, SSE or WebSocket, and local ones run as stdio processes. One example of what a connected server can do: act as a channel that pushes Telegram, Discord or webhook events into a session while you’re away.
Anthropic is asking plugin builders to add Tasks and Elicitations to their MCP servers now. The server-side guides are in the docs under MCP tasks for server authors and responding to MCP elicitation requests.
Related stories
- Claude plugins get a portal as MCP usage climbs 110x
- Anthropic hands Agent Skills to the foundation behind MCP
- OAuth login to MCP servers is fixed
- Claude Code 2.1.309 closes sandbox and permission holes
- Claude apps switch to auto mode by default next week
- Claude Code 2.1.307 moves API code to version 2026-10-21
Comments
No comments yet. Be the first.
Join the conversation
Sign in to leave a comment. It appears after moderation.
