Verified65% confidenceFactExact time
The MCP protocol itself does not carry risk semantics for tools, and the responsibility for security judgment falls back on the application developers.
3
Sources
65%
Confidence
Long-term
Relevance
7/14/2026
First Seen
Sources
AI Agent人在回路（HITL）设计：风险分级与最佳实践
Related Claims
UnverifiedMCP is permission-based, and software companies only expose the features they are willing to open up, meaning commercially sensitive or core monetization features will almost never be included.78% similarUnverifiedThe security gateway is designed to be non-invasive, requiring developers to neither rewrite existing Agent logic nor MCP implementations, but only to deploy the gateway on the communication link.78% similarUnverifiedMCP only addresses the standardization of connectivity and invocation, and is not responsible for security; responsibilities such as authentication and authorization, approval of dangerous operations, and verification of service trustworthiness must be handled by the application layer.77% similarUnverifiedMCP's capability discovery does not mean the model gains unrestricted access; authentication, permission, and approval mechanisms remain indispensable.76% similarVerifiedMCP currently remains focused on standardizing tool discovery and invocation formats, while fine-grained permission management control has not yet been incorporated into the core protocol constraints.76% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/504475API
curl https://kongchang.com/api/v1/knowledge/claims/504475MCP
get_claim(id=504475)