mcp-server-management

Warn

Audited by Gen Agent Trust Hub on Apr 6, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: Sensitive credentials including GitHub tokens and API keys are passed as plain-text command-line arguments to the setup-mcp-server.sh and setup-mcp-proxy.sh scripts, making them visible in the process list and shell history.
  • [DATA_EXFILTRATION]: The scripts access sensitive files on the local system, including session cookies via HIGRESS_COOKIE_FILE and worker gateway keys stored in /data/worker-creds/, to facilitate gateway configuration.
  • [COMMAND_EXECUTION]: The setup-mcp-server.sh script uses sed to inject credentials into YAML templates without sufficient validation, which could be exploited for configuration manipulation if input values are not strictly controlled.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through its ingestion of user-provided configuration files. Ingestion points: The --yaml-file argument in setup-mcp-server.sh. Boundary markers: No delimiters or ignore instructions are applied to the user-provided YAML content. Capability inventory: The skill can update gateway settings, write to local agent configuration files, and upload data to MinIO storage. Sanitization: Employs simple string substitution for credentials without robust validation of the surrounding YAML structure or content.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Apr 6, 2026, 08:12 AM