WebMCP Tool Safety Annotations
Configures safety metadata for WebMCP tools to manage browser permission prompts and optimize agent execution behaviors.
This skill enables developers to implement WebMCP tool annotations—specifically readOnlyHint, destructiveHint, and idempotentHint—which are essential for securing agentic interactions within the browser. By providing these hints, the skill helps determine when a browser should automatically approve a tool call or require explicit user consent, ensuring a balance between seamless automation and robust security. It is particularly valuable for commerce applications where actions like 'Add to Cart' versus 'Place Order' require different levels of verification and safety handling.
主要功能
01Standardizes browser permission prompt logic via WebMCP metadata
02Configures destructiveHint for irreversible or high-risk actions requiring consent
03Applies idempotentHint to signal safe retry capabilities to the agent
0432 GitHub stars
05Implements readOnlyHint for non-state-modifying tools to streamline UX
06Provides pre-defined annotation mappings for commerce-specific tools
使用场景
01Implementing safety-first browser agents that enforce manual consent for destructive actions
02Optimizing agent UX by auto-approving safe, read-only search and fetch queries
03Building secure checkout flows for agentic e-commerce applications