openai/tunnel-client

Customer-run client for Secure MCP Tunnel: connect private or localhost MCP servers to ChatGPT, Codex, the Responses API, and AgentKit without exposing them to the public internet.

What it solves

tunnel-client allows users to connect private or localhost MCP (Model Context Protocol) servers to OpenAI-hosted products like ChatGPT, Codex, and the Responses API without exposing the MCP server to the public internet. It bypasses the need for new inbound firewall rules or public endpoints, making it suitable for servers running on laptops, VMs, or private networks.

How it works

The project provides a customer-run agent that establishes a secure tunnel to an OpenAI-hosted MCP tunnel endpoint. It can be run as a standalone daemon with operational monitoring (health, readiness, and metrics) or embedded directly into a Go process as an SDK. For deployments, it supports various environments including Docker, Kubernetes, and VMs, and offers a Cloudflare companion for managed provisioning.

Who it’s for

Developers and enterprise users who have MCP servers in private environments and need to securely connect them to OpenAI's AI tools while maintaining strict network security boundaries.

Highlights

  • Secure Connectivity: Connects local MCP servers to ChatGPT and Codex without public internet exposure.
  • Flexible Deployment: Supports standalone binary, Docker, Kubernetes, VM, and Go SDK embedding.
  • Operational Visibility: Includes built-in /healthz, /readyz, /metrics, and /ui endpoints for monitoring.
  • Enterprise-Ready: Provides detailed SBOMs, vulnerability reports, and Sigstore provenance for secure supply chain validation.

Related

  • Project
    Waishnav/devspace

    A self-hosted MCP server that gives ChatGPT secure access to your local files and terminal to read, edit, and run code directly in your projects.

  • Project
    rebel0789/codexpro

    A local MCP server that connects ChatGPT to your local machine and allowed repositories, enabling the AI to read, search, and edit code directly.

  • Project
    zerodha/kite-mcp-server

    An MCP server that gives AI assistants secure access to the Kite Connect trading API for market data retrieval, portfolio management, and trade execution.

  • Project
    agentclientprotocol/codex-acp

    An ACP adapter for OpenAI Codex that allows ACP-compatible clients to interact with Codex as an agent server for AI-assisted coding and system operations.

  • Project
    googleapis/mcp-toolbox

    An open-source MCP server that connects AI agents and IDEs to enterprise databases through prebuilt generic tools or a customizable framework for secure, structured data access.