setup-mcp
Fail
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: HIGHREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill provides commands to install the Deepgram CLI by piping shell scripts from https://deepgram.com/install.sh to sh and downloading PowerShell scripts from https://deepgram.com/install.ps1 for execution via iex.
- [EXTERNAL_DOWNLOADS]: The instructions guide the user to download software from several sources including the official Deepgram GitHub repositories, Homebrew taps, and the Python Package Index (PyPI).
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates the integration of an MCP server that retrieves information from external documentation and knowledge sources at runtime. This behavior creates a surface where external content might influence the AI agent's instructions.
- Ingestion points: Data retrieved from https://developers.deepgram.com/_mcp/server and other Deepgram knowledge retrieval tools.
- Boundary markers: Not present in the setup instructions.
- Capability inventory: The configured tools allow the agent to perform knowledge searches and network requests to Deepgram APIs.
- Sanitization: Not explicitly addressed in the installation steps.
- [DYNAMIC_EXECUTION]: The skill describes functionality where the MCP server dynamically fetches its available tool list from the vendor's API at runtime.
Recommendations
- HIGH: Downloads and executes remote code from: https://deepgram.com/install.sh - DO NOT USE without thorough review
Audit Metadata