mcp-cli

The CLI itself is designed for configuration management and does not directly execute user-defined commands from `mcp-compose.yml` in a shell (e.g., no `eval`). Instead, it parses them into structured `command` and `args` fields in a JSON output file, which is then consumed by other AI tools. Remote server authentication uses the OAuth 2.0 client credentials flow, acquiring an access token that is then used in the output configuration, rather than storing raw client secrets directly. Environment variables for sensitive data are encouraged (e.g., API keys, OAuth secrets) and handled through `.env` files or system environment. The main security considerations would be the user's trust in the configured `mcp-compose.yml` content (e.g., malicious `command` values or OAuth `token-endpoint` URLs configured by the user) and the security practices of the downstream AI tools that execute the generated MCP configurations.