Learn

MCP security concepts

Reference explainers on the attack patterns aimcplist grades for — what they are, how they actually happen, and how to check for them yourself.

What is MCP tool poisoning?

A tool's own description can carry instructions a user never sees — and an AI model reads it anyway.

The tool you approved isn't always the tool you're still running.

A tool that fetches a URL on your behalf can be tricked into fetching the wrong one.

A server that still installs fine can still be running code nobody's maintaining anymore.

The specific things to actually check before you let a model call an MCP server's tools.

The individual-developer checklist isn't enough once a tool is running against production systems and real customer data.