Secure SQL AI Analyst: Enabled by Pomerium & ChatGPT Developer Mode

See how to connect a custom SQL data to ChatGPT Dev Mode via Pomerium's agentic access management to create an AI business analyst.

September 12, 2025

OpenAI’s new ChatGPT Developer Mode adds native support for Model Context Protocol (MCP) servers. For developers building agent tools, this is a big step: you can now plug in custom MCP servers, like a SQLite database, directly into ChatGPT. However, bridging local development environments with cloud-hosted LLMs isn’t simple, and VPNs don’t work in this model. That’s where Pomerium comes in.

The demo below shows how to make bridging local MCP servers with a cloud hosted LLM like ChatGPT Developer Mode secure, simple, and scalable. And it all starts with choosing the right MCP gateway.

Got Ideas?

Ideas for how to make this better? We'd love to hear them. Some things we're thinking about is how to connect multiple clients and make it easier to jump between them all within the same UX. Connect with us on social or mcp@pomerium.com.

How it works

Watch video

Video Recap

Connecting MCP Server to ChatGPT Dev Mode:

Open up a simple SQLite MCP server on a laptop.
Using Pomerium’s identity-aware proxy, they created a secure reverse tunnel.
That tunnel terminated in Pomerium, which handled auth, TLS certificates, and per-request policy enforcement.
From there, ChatGPT Dev Mode could talk to the local MCP server as if it were cloud-hosted.

Once the tunnel was set up:

ChatGPT listed database tables via MCP tools.
It generated SQL queries automatically, including a sales-by-country analysis.
Code Interpreter rendered a heatmap, then followed up with product trend queries.
The entire workflow took minutes, no manual SQL needed.

Why It Matters:

Security: VPNs are useless here — hosted LLMs won’t sit behind your corporate VPN. Pomerium enforces Zero-Trust, per-request policies at Layer 7.
DX Win: OAuth flows, cert management, DNS, and upstream tokens are handled by config in Pomerium, not custom MCP server code.
Flexibility: Developers can build locally, test securely, and share with peers using the same route.

Broader Implications:

Teams can expose dev databases or metrics sources safely for AI agent loops.
Non-technical users could consume MCP-based reports through ChatGPT with secure upstream credentials.
Enterprises get guardrails from day one of MCP development.