Resources/AI Security & Hacks Library/Postmark MCP supply-chain trojan with silent BCC exfiltration
incidentHighSeptember 2025Multi-source3 references

Postmark MCP supply-chain trojan with silent BCC exfiltration

A trojanized Postmark MCP package on npm silently BCC'd every outbound email to an attacker-controlled inbox.

Skill or Plugin BackdoorsData Exfiltration

Affected systems

MCP deployments, Long-lived agents, Customer support agents

Primary threats

Skill or Plugin Backdoors, Data Exfiltration

Impact types

Silent data exfiltration, Connector impersonation

CVEs

Not specified

What an auditor should now check

  • Verify connector provenance, update flow, and package ownership history
  • Inspect whether outbound channels can be silently copied or retargeted
  • Check whether email or messaging tools expose destination-level approval and audit trails

Why this matters

This is the cleanest example of MCP impersonation and agentic supply-chain failure. The connector looked legitimate while exfiltrating sensitive communication.

What happened

A trojanized package impersonated the Postmark MCP connector and silently copied outbound email traffic to the attacker. The connector retained the user's normal authority and looked legitimate enough to be trusted.

Why the classification matters

This is not just a package-integrity problem. It is a live example of how connector trust collapses into operator harm once agents hold real-world communication authority.

What an auditor should now check

  • Whether outbound connectors can silently add recipients or mirrors
  • Whether connector identity and provenance are pinned and reviewed
  • Whether message destinations and payloads are logged before dispatch

Zealynx takeaway

If a connector can reach customers, counterparties, or internal inboxes, it needs the same scrutiny as a privileged SaaS integration.

Control implications

  • Require provenance checks and pinning for connectors with outbound communication authority
  • Log full connector descriptors and outbound destinations
  • Do not treat vendor-like package names as evidence of trust

Affected systems

  • MCP deployments
  • Long-lived agents
  • Customer support agents

Impact types

  • Silent data exfiltration
  • Connector impersonation

Smart Contract Security Digest

Monthly exploit breakdowns, audit checklists, and DeFi security research — straight to your inbox

© 2026 Zealynx