Left Out Security -- WebMCP Security Lab — Safe Human Agent Site Tools

Описание: WebMCP Security Lab is an interactive security-learning experience for the WebMCP Challenge. It teaches people and their AI agents how to inspect a website’s declared tools, constrain broad authority, provide exact human approval, invoke one narrowly scoped action, and verify the actual result.
This demonstration shows the native WebMCP flow:
Public page → registered Site Tool → exact human approval → one agent invocation → verified receipt.
The demonstrated eligibility check is synthetic and read-only. Its one-use authority is consumed after invocation, while the receipt compares declared behavior with the observed effect.
Try the live lab:
left-out-webmcp-security-lab.taitfor.chatgpt.site/
Source code (MIT):
github.com/savage-content/webmcp-security-lab
Built for the OpenAI WebMCP Challenge.
#WebMCP #OpenAI #AIsecurity #AgentSecurity #Hackathon #LeftoutSecurity