Executive brief
@translated/lara-mcp is an MCP Server that integrates with the Lara Translate API for translation memory management. A command injection vulnerability in the import_tmx tool allows attackers to execute arbitrary system commands on the host running the server, potentially through indirect prompt injection when the MCP client processes untrusted content.
Technical details
The vulnerability is a command injection flaw in the import_tmx function, which uses Node.js child_process.exec() to execute curl commands with unsanitized user input from the tmx_url parameter. An attacker can inject shell metacharacters (such as $(), |, &&) into the tmx_url field to execute arbitrary commands under the server process's privileges. The attack can be triggered directly via MCP tool invocation or indirectly through prompt injection when an MCP client processes malicious instructions embedded in untrusted files or remote data. No authentication is required to invoke the tool. The advisory recommends using child_process.execFile instead of exec, which prevents shell metacharacter interpretation.
Affected products
- @translated lara-mcp <=0.0.11
Timeline
- 2025-07-21: disclosed
- 2025-07-21: advisory