LayerX — Claude Desktop Extensions zero-click RCE via calendar event

• Category: Security

AI relevance: This is a tool-chain vulnerability in an LLM agent ecosystem (MCP/Claude Desktop Extensions) that can convert untrusted calendar data into local code execution.

  • LayerX reports a zero-click RCE chain where a Google Calendar event can trigger local code execution via Claude Desktop Extensions.
  • Desktop Extensions are MCP servers packaged as .mcpb bundles that run on the host machine.
  • Unlike browser extensions, these connectors execute unsandboxed with full system privileges.
  • The attack hinges on autonomous tool chaining: a low-risk connector (Calendar) feeds data into a high-risk local executor.
  • LayerX says the issue impacts 10,000+ active users and 50+ extensions.
  • The report assigns CVSS 10.0 and notes the architectural trust-boundary problem remains unresolved.

Why it matters

  • Agentic desktop tooling can turn benign data sources into privileged execution paths.
  • “Low-risk” connectors become high-risk when they can silently invoke local tools.
  • It underscores the need for hard boundaries between untrusted context and local execution.

What to do

  • Gate privileged tools behind explicit, user-visible confirmation.
  • Sandbox MCP connectors or run them with least-privilege OS accounts.
  • Audit tool-chaining policies so untrusted inputs cannot reach local executors.

Links