MCP Tool Annotation Support — Now Generally Available

Builders can label and annotate MCP tools so AI clients can distinguish routine actions from sensitive ones — keeping approval prompts meaningful and reducing the “approve everything” reflex.

• Editable Tool Titles — Relabel a tool’s display name without breaking clients already invoking it by its programmatic name.
• Behavior Hints — Set destructiveHint, readonlyHint, idempotentHint, and openWorldHint on each tool so AI clients know how a tool behaves, not just what it’s called.
• Automated Control — Read-only tools can be set to always allow; write and delete actions still require explicit approval.
• Reduced Approval Fatigue — Routine lookups execute without interruption, so approval prompts surface only when they actually matter.

Learn more about Tool Annotations.