A malicious update in the npm package postmark-mcp introduced a backdoor that silently exfiltrates emails from users to an external server, highlighting severe vulnerabilities in the trust model of MCP servers used by AI assistants. With over 1,500 weekly downloads, developers unknowingly handed over complete email control to a compromised tool, raising alarms about the security of tools integrated into enterprise workflows. Immediate action is required to remove the malicious package and audit other MCP servers for similar risks.