Secure MCP Tunnel

Secure MCP Tunnel is a solution that connects private MCP (Model Context Protocol) servers to the OpenAI product suite without opening inbound firewall ports or exposing the server to the public internet. It works by having a tunnel-client running on the internal network create an outbound HTTPS path to OpenAI, which fetches pending tasks, forwards the requests locally, and returns the responses.

This technology is especially useful in environments where MCP servers reside in private networks, on-premises, or behind firewalls. It helps supported OpenAI services such as ChatGPT, Codex, and the Responses API securely invoke the capabilities of local servers while maintaining security boundaries.

The process works as follows:

• Create an OpenAI-hosted MCP tunnel endpoint in the platform settings.
• Run the tunnel-client on the internal network and configure the tunnel ID and the private MCP server address.
• The tunnel-client polls OpenAI to fetch pending JSON-RPC requests, forwards them to the local server, and sends the responses back through the tunnel.

On the network side, no separate inbound internet access is required. Instead, only an outbound HTTPS connection to api.openai.com:443 (or mtls.api.openai.com:443 when mTLS is configured) and access to the local MCP server are needed.