Claude Code 2.1.76: MCP Elicitation and New Hooks Revolutionize AI Agents

Anthropic has released Claude Code version 2.1.76, a significant update introducing MCP (Model Context Protocol) Elicitation support and a range of new hooks that fundamentally change how AI agents interact with users and systems.

MCP Elicitation: Structured Input Mid-Task

The headline feature is that MCP servers can now request structured input in the middle of an ongoing task via an interactive dialog. This means AI agents can pause, ask the user for specific information through form fields or a browser URL, then continue — all without losing context.

For CIOs and developers building agentic workflows, this represents a quantum leap: agents can now handle complex scenarios requiring human input mid-process without having to restart from scratch.

New Elicitation Hooks Give Full Control

The new Elicitation and ElicitationResult hooks let developers intercept and override responses before they're sent back to the MCP server. This opens up sophisticated validation, logging, and security control of all input passing through the agent.

Additional Updates

• -n / --name CLI flag – Set a display name for the session at startup, useful for multi-agent orchestration
• worktree.sparsePaths – Large monorepo? Check out only the directories you need via git sparse-checkout
• PostCompact hook – Fires after compaction completes, ideal for state synchronization
• /effort command – Set the model's effort level directly from the terminal
• Session quality survey – Enterprise admins can configure the feedback rate via feedbackSurveyRate

Relevance for Enterprise Leaders

For organizations building on Claude Code or integrating AI agents into existing workflows, MCP Elicitation enables intelligent processes that can actually pause and ask for help when needed — rather than guessing or failing.

The update is available now via npm.