RelayShield's Agentic Attack Surface Is Now an MCP Server

AI agents get compromised differently than humans do. A typosquatted MCP server, a prompt-injection attack that exfiltrates credentials, an agent framework sitting on an unpatched CVE, a service-accou