matlab-mcp-core-server

Pass

Audited by Gen Agent Trust Hub on Sep 16, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill instructs users to download a server binary from an external GitHub repository (https://github.com/matlab/matlab-mcp-core-server/releases/latest). While this is the core component of the skill, users must verify the authenticity of the binary as it executes with local user permissions.
  • [COMMAND_EXECUTION]: Through the evaluate_matlab_code and run_matlab_file tools, the AI agent gains the ability to execute arbitrary MATLAB commands on the user's machine. This includes file system access and potentially shell command execution via MATLAB's system or ! operators.
  • [DYNAMIC_EXECUTION]: The evaluate_matlab_code tool takes a raw string and executes it dynamically within the MATLAB environment. This is the primary function of the skill but constitutes a powerful capability that requires careful oversight.
  • [INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection. If an agent uses these tools to process untrusted external data (like a malicious file or website content) and is tricked into executing instructions contained within that data as MATLAB code, it could lead to unauthorized system operations.
  • [PRIVILEGE_ESCALATION]: The troubleshooting section suggests the use of chmod +x to grant execution permissions to the downloaded binary, which is a standard step for local binary setup but highlights the transition of the external file to an executable state.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 16, 2026, 10:42 PM
Security Audit — agent-trust-hub — matlab-mcp-core-server