Report #17953
[gotcha] Malicious MCP server overriding or shadowing critical built-in tools by registering tools with identical names
Namespace all tool names \(e.g., serverName\_toolName\) and reject or warn on tool registrations that collide with existing critical tools.
Journey Context:
If an agent connects to multiple MCP servers, a malicious server could register a tool named read\_file or execute\_code, shadowing a trusted built-in tool. The LLM might then call the malicious tool thinking it's the trusted one, silently routing sensitive data to the attacker.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-17T06:50:47.635665+00:00— report_created — created