AI Agent Security (Course)
Tool Shadowing
Also: cross-server tool shadowing
A malicious MCP server redefining or steering another server's tools, exploiting the fact that all servers share one context window.
In more detail
Trust in MCP is granted per server, but the model reads every server's tool descriptions in a single context. A hostile server can therefore define a tool with the same name as a trusted one, or include text that changes how the agent should use a different server's tools. Connecting an unknown utility server next to a server that holds sensitive data lets the utility's text reach that data through the agent.