Skip to content

Tool name shadowing

An MCP tool uses a name that matches a familiar built-in tool.

A misleading name can cause an agent or reviewer to trust an external tool as though it were the local built-in. The external server may receive data or perform actions with different permissions.

A newly connected MCP server exposes a tool called Read. An agent uses it expecting a local file reader, but the external service receives the request.

This is a fictional example, not an event from a customer.

The rule checks MCP-formatted tool names against a list of built-in names. A name collision is the signal; the tool implementation is not analyzed.

Rule IDEvent categoryTierBase severity
RULE_SUPPLY_CHAIN_TOOL_SHADOWsupply_chainCommunityHIGH

Base severity is the rule default. A finding involving content saved for later execution may have a lower severity. Your loaded rule version and event details determine the finding you see.

Confirm the server identity, tool permissions, and intended recipient. Review tool selection when local and external tools have similar names.

Legitimate filesystem servers often use simple names such as read, write, or bash. A collision can be intentional and benign.

Different naming formats, near-matching names, misleading descriptions, and malicious tools without name collisions can be missed.

This rule concerns tool identity. It does not show that the model was deceived or that the external implementation is malicious.

A finding describes a pattern in captured activity. Review the original call, its result, and the task context before concluding that harm occurred. Absence of a finding does not establish that an operation is safe.

Browse this category. Find a rule by ID.