A guide · ~6 min read
ChatGPT Developer mode, explained
Developer mode turns ChatGPT into a full Model Context Protocol client. It can connect to any remote MCP server you point it at and call both read and write tools. It is the most powerful setting in ChatGPT, and the one OpenAI attaches the loudest warning to.
How to enable it
- Open ChatGPT on the web. There is no mobile equivalent.
- Go to Settings → Security and login.
- Turn on Developer mode.
- The apps and connectors page will now offer developer-mode apps from the plus button.
Not seeing it? Check the plan — OpenAI lists Pro, Plus, Business, Enterprise and Education — and check whether you are in a managed workspace where an admin has switched it off.
The risks, stated plainly
OpenAI names three, and none of them are theoretical:
- Prompt injection. Content the model reads can contain text designed to look like instructions. A web page or a document can try to talk your assistant into doing something you did not ask for.
- Model mistakes on writes. The model can call the right tool with the wrong arguments. On a delete operation that is not recoverable.
- Malicious MCP servers. A server you connect can attempt to extract information from the conversation. There is no review process standing between you and an arbitrary URL.
Practical posture: connect servers you can identify the operator of, prefer read-only tokens where a server offers them, leave write confirmation on, and actually read the payload before approving.