OpenAI's move has clarified enterprise internal network security: one-way outbound + whitelist authorization, finally eliminating the need to choose between convenience and exposure.

OpenAI launches a secure MCP tunnel, allowing enterprises to connect internal network tools to ChatGPT, Codex, and Responses API without exposing IP addresses or opening inbound ports. By establishing a unidirectional outbound HTTPS channel with a local tunnel-client, cloud models can access internal networks without crossing boundaries. Enterprises retain control, achieving precise authorization through interface whitelists and workspace role bindings, compatible with custom certificates, outbound proxies, and mTLS. The open-source client supports Kubernetes Sidecar, Pods, system services, and provides a local management interface.