Vulnerability  ·  2026-09-18

cc-connect Feishu MCP agent connector — per-user allowlist bypass in interactive card callback handler

VulnerabilityHigh impactGlobalCVE-2026-92801
NVD published this CVSS 8.8 High vulnerability on Sept 16, 2026 describing an authorization-consistency gap between two input channels of the same chat-to-agent bridge.
This is a small but instructive example of the broader agentic pattern noted across the MCP ecosystem this week: authorization checks applied inconsistently across multiple input surfaces of the same agent-command bridge, allowing an attacker who only has access to one input channel to command a privileged agent through another.
cc-connect enforces per-user allowlist filtering in its text-message handler but not in the onCardAction handler for Feishu interactive card callbacks, letting an attacker dispatch agent commands by triggering card actions in chats the attacker has been admitted to, bypassing the access controls meant to restrict who can command the agent.
cc-connect <= 1.5.0
Upgrade cc-connect beyond 1.5.0 once a fix is released; apply consistent authorization checks across all input handlers (text and card-action) that can trigger agent commands.
cc-connect GitHub
See this in the live feed Explore related AI security and governance findings — updated every morning.
Open the feed →