ali-opensendbox-mcp-server

The server correctly implements an MCP interface for OpenSandbox API calls. No direct use of `eval` or obvious obfuscation. Hardcoded secrets are not present; API key is configured via environment variable. Network interactions involve dynamically retrieving an 'embedding-proxy-port' from sandbox metadata for execd operations, which is then used as a temporary base URL for command and file actions. This relies on the OpenSandbox system itself being secure and the metadata being trustworthy. The primary security consideration is that the 'command_run' tool allows arbitrary shell command execution within a sandboxed environment; if an untrusted AI or user has access to this tool, they could potentially execute harmful commands inside the sandbox. The server itself appears robust against external injection and generally follows good practices for handling external API interactions.