Home/AI Incident Database/Remote code execution in an AI IDE through prompt injection into an auto-approved MCP server
Remote code execution in an AI IDE through prompt injection into an auto-approved MCP server
Researchers showed that a prompt injection reaching Cursor's agent through a connected MCP server could write to the project's MCP configuration, which the IDE auto-executed without confirmation, giving an attacker code execution on the developer's machine.
Contribute
Spot an error, or know a record we should add? Members can propose additions and submit corrections.