native-mcp
Pass
Audited by Gen Agent Trust Hub on May 19, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONREMOTE_CODE_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill describes using standard package managers like
npx(Node.js) anduvx(Python) to fetch and execute MCP servers from public registries. This is the primary intended mechanism for extending the agent's capabilities via the Model Context Protocol. - [COMMAND_EXECUTION]: The stdio transport mechanism executes local subprocesses based on user-defined configurations in
~/.hermes/config.yaml. The skill documentation explicitly details security mitigations, such as filtering the inherited environment variables to prevent accidental exposure of sensitive keys likePATHandHOME. - [REMOTE_CODE_EXECUTION]: The HTTP transport configuration allows the agent to connect to remote MCP servers. These servers can provide tool definitions and execute logic externally, returning results to the agent. This behavior is a core feature of the protocol and is managed through explicit user configuration.
- [SAFE]: The skill documentation emphasizes security best practices, including a 'filtered' environment for subprocesses and a redaction system that strips API keys (GitHub, OpenAI, etc.) and bearer tokens from error messages before they are processed by the LLM.
Audit Metadata