connect-mcp-server
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONCREDENTIALS_UNSAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill documents the use of the
claude mcp addcommand to integrate local and remote services. It provides various examples of shell commands including local executable paths and system wrappers likecmd /cfor Windows compatibility. - [EXTERNAL_DOWNLOADS]: The documentation references multiple MCP server packages available via
npx, such as@modelcontextprotocol/server-filesystem,@modelcontextprotocol/server-github, and@modelcontextprotocol/server-postgres. These are official or well-known community packages used to extend agent capabilities. - [INDIRECT_PROMPT_INJECTION]: The skill includes a dedicated security section warning that servers fetching untrusted content (e.g., web search results or database records containing external data) can expose the agent to prompt injection. It recommends mitigations such as limiting permissions and using trusted sources.
- [CREDENTIALS_UNSAFE]: The guide explicitly advises against hardcoding credentials in configuration files. It provides examples of using environment variable expansion (e.g.,
${GITHUB_TOKEN}) to manage secrets securely and recommends the use of OAuth 2.0 where available.
Audit Metadata