Which MCP threats should a gateway stop?

In short: these eight pages describe the attacks a gateway is meant to stop, from poisoned tool descriptions to tokens leaking out of config files. Each page explains what happens, how a gateway can mitigate it and which scored gateways document a mitigation. Sources are the MCP and OWASP pages we read.

• What is tool poisoning?
• What is a rug pull in MCP?
• What is the confused deputy problem in MCP?
• Why is token passthrough a threat?
• What goes wrong with over-broad MCP scopes?
• What are shadow MCP servers?
• How do credentials leak from MCP configs?
• How does prompt injection arrive through tool output?

By The Datasheet desk · Published and reviewed 2026-11-03