Systeme Growth

Systeme Growth guide

Connecting Systeme.io to ChatGPT: A Working Setup

Systeme.io documents ChatGPT setup, but the page predates the server. Here is the working setup, the plan and platform limits, and why you pick No Auth for a server that now asks for OAuth.

Audit your page free

Systeme.io publishes a setup guide for ChatGPT, which is more than it does for any other assistant. Follow it exactly and the connection works. What the guide does not tell you is that the page was last updated on 8 June 2026 and the server has moved since.

Three things are worth knowing before you start, because each one costs time to discover on your own: developer mode is required and it is a web-only feature, the settings path in the guide does not match the one OpenAI documents, and the live server now answers with an OAuth challenge that the documented setup does not mention.

None of them stops you. All three change what you do.

Watch instead: How to connect your chagpt to systeme io via mcp to create and sell digital product

Check that your account can do this at all

Custom MCP connectors live behind developer mode, and OpenAI's own documentation states it is "Available to Pro, Plus, Business, Enterprise, and Education accounts on the web."

Read that last phrase carefully. On the web. The phone app is not a route to this, no matter which plan you pay for. If you were planning to run your funnel from your phone, that is the detail to absorb before anything else.

This is also the one point where the two vendors' documentation disagrees about where to click. Systeme.io's guide sends you to Settings, then Apps and Connectors, then Advanced settings. OpenAI documents Settings, then Security and login. Interfaces move; when the instructions differ, follow the vendor whose interface you are actually standing in.

Get the key

In your Systeme.io dashboard, open profile settings and find "MCP & API keys". Generate a key, name it something you will recognise later, and choose an expiry.

The key is the credential. In this setup it travels in the connector URL rather than in a header, which means it will sit in a settings field in ChatGPT and in your clipboard history. Treat it the way you would treat a password, and never let it appear on screen while recording.

Add it to ChatGPT

With developer mode on, create a new app or connector. Name it Systeme.io. The server URL is the MCP endpoint with your key attached as a query parameter:

  • Server URL: https://mcp.systeme.io/mcp?mcpKey=YOUR_KEY
  • Authentication: No Auth — the key is already in the URL
  • Transport: ChatGPT supports SSE and streaming HTTP; the Systeme.io endpoint is a streaming HTTP server

Acknowledge the warning, create it, and open the app's details page to see the tools it pulled in. That list is the real answer to what the connection can do — not the documentation, and not this article.

Why you pick "No Auth" for a server that asks for OAuth

This is the part that looks wrong and is not. Send an unauthenticated request to the endpoint today and it does not simply refuse. It replies 401 with an authentication challenge naming a scope of mcp:full and pointing at its own OAuth metadata.

Follow that metadata and there is a complete, working OAuth setup behind it: an authorisation server, an authorisation endpoint inside the Systeme.io dashboard, a token endpoint, and PKCE support. A reader who has just been told to select "No Auth" is entitled to wonder whether the guide is simply out of date.

It is out of date, but not in a way that changes your answer. ChatGPT's OAuth option expects to register itself with the authorisation server automatically, through dynamic client registration or a client ID metadata document. The Systeme.io authorisation server advertises neither. It lists no registration endpoint, and it explicitly reports that client ID metadata documents are not supported.

So the OAuth door exists and is genuinely open, but there is no automatic way for ChatGPT to let itself in. The key in the URL is the path that works today.

Two doors into the same server, and only one you can walk through today

Key in the URL — use this

  • Select "No Auth" in ChatGPT
  • The key travels as the mcpKey query parameter
  • Works today, exactly as documented
  • Cost: the credential lives in a URL, so rotate it on a schedule

OAuth — advertised, not yet reachable

  • The live server answers 401 with a scope of mcp:full
  • It publishes an authorisation endpoint, a token endpoint and PKCE
  • But no registration endpoint is advertised
  • And client ID metadata documents are reported unsupported
  • So ChatGPT cannot register itself automatically
Both routes are real. Pick the documented one until the authorisation server advertises a way for a client to register itself.

What it can do once connected

The documented surface is contacts, tags, tag assignment, contact fields and newsletters — list, retrieve, create, update, delete.

The demonstrated surface is larger. Creating digital products, building offers and creating funnels have all been run against a live account, on camera, even though the developer documentation does not list them. The documentation describes an earlier version of the server than the one you are connecting to.

That gap cuts both ways. Capability you were told you did not have may be there, and capability you saw working may move without an announcement. Check the tool list in the app's details page rather than trusting any written list, including this one.

Two habits worth keeping

Leave the confirmation prompts on. OpenAI's documentation is direct about it: "Write actions by default require confirmation. Carefully review the tool input which will be sent to a write action to ensure the behavior is as desired." A create or delete that reads reasonably in a chat message can still be wrong about which contact or which list it lands on.

Rotate the key on a schedule rather than after an incident. A credential in a URL is easier to leak than one in a header — into a screenshot, a support thread, a recording, a browser history. Set an expiry when you generate it, and replace it before it lapses rather than discovering it lapsed mid-task.

Worth saying plainly: the assistant can build a funnel by conversation in less time than it takes to open the page builder, and that changes nothing about whether the funnel converts. Speed of assembly is not quality of offer. Audit the page your traffic actually lands on, fix what the evidence names, and let the assistant handle the list behind it.

Technical disclosure

Two kinds of evidence sit behind this piece. The setup steps, the connector URL and the "No Auth" instruction were read from Systeme.io's own ChatGPT integration guide on 27 September 2026, where the page carried a last-updated date of 8 June 2026. The plan availability, the settings path, the supported transports, the authentication options and the write-confirmation wording come from OpenAI's developer-mode documentation, read the same day. The OAuth challenge, the scope, the authorisation and token endpoints, the PKCE support and the absence of a registration endpoint were observed directly from the live server and its published metadata on that date, not from any documentation page. Where a document and the server disagree, the server is the authority here. Capability on a service still in beta moves without announcement, so test before depending on anything, including this.

Audit your page free, then come back to the list.