A guide · ~7 min read
ChatGPT MCP: giving ChatGPT a memory it does not own
ChatGPT supports remote MCP servers. That is a bigger deal than it sounds, because it means the model no longer has to remember you inside its own small memory bucket — it can look you up, live, from somewhere you control.
What MCP actually is
The Model Context Protocol is an open standard for connecting AI assistants to external tools and data. A server exposes a set of tools; a client — ChatGPT, Claude, Cursor, others — discovers those tools and calls them mid-conversation when they are useful.
Crucially it is not an OpenAI feature. It is a protocol. The same server answers Claude and ChatGPT identically, which is why it is the right foundation for anything you want to survive a change of provider.
Who can use it
ChatGPT's MCP support lives behind Developer mode. Per OpenAI's documentation it is available to Pro, Plus, Business, Enterprise, and Education accounts, on the web only. There is no mobile support.
OpenAI labels Developer mode as elevated risk, and they are right to. It grants full read and write tool access. Connect servers you trust, read write actions before confirming them, and treat an unfamiliar MCP URL the way you would treat an unfamiliar browser extension.
Connecting Konshus
Konshus runs a remote MCP server over streaming HTTP with OAuth 2.1. The URL is:
https://konshus.ai/api/mcp
- In ChatGPT, open Settings → Security and login and turn on Developer mode.
- Go to the apps and connectors page and add a new developer-mode app.
- Paste the URL above.
- Sign in to Konshus when prompted and approve access.
- In a conversation, pick Developer mode from the composer and select Konshus.
Step-by-step with screenshots: how to add an MCP connector to ChatGPT.
What it changes
ChatGPT's own memory holds a paragraph. A connected vault holds years. Instead of cramming a summary of you into eight slots, the model queries for what it needs when it needs it — the relevant project, the relevant preference, the thing you said in March.
And because the vault sits outside the provider, the same context answers Claude tomorrow. Switching models stops costing you your history. That is the argument in the hidden cost of switching AI models, finally with a technical answer attached.