ChatGPT reaches Develosaur two different ways, and they answer two different questions: "let me try this right now" and "let this run against my real account." Here is the full path for both.
Sandbox mode: nothing to set up
If ChatGPT can fetch a URL, you do not need an account, a connector, or a setting changed anywhere. Paste this into the chat:
Read https://mcp.develosaur.com/llms.txt and use it to build me a develosaur board for my project
ChatGPT fetches that file, which is written for an agent rather than a person, follows it, and builds a temporary board. You get a link back to watch the graph as it is built, live. The board is throwaway by design, kept only for a short window, until you sign up and claim it as a permanent project. Nothing was configured, nothing was installed, and nothing needed an OAuth screen at all.
Connector mode: your real boards
For a board tied to your account that persists on its own, ChatGPT calls this Developer mode, and it is a real, signed-in MCP connection, not a limited version of one.
- On chatgpt.com, open Settings, then Security and login, and turn on Developer mode. It is available on Plus, Pro, Business, Enterprise and Education.
- In the apps section, choose the option to create a developer-mode app, paste
https://mcp.develosaur.com/v1/user/mcpas the address, and choose OAuth. - Finish the sign-in with your existing account, then pick the app from the Developer mode tool in the composer.
That address is the same one every other client uses, and the sign-in is the same browser consent flow described in why Develosaur has no API keys to paste. There is nothing ChatGPT-specific to configure beyond turning the feature on; the connector talks to the same board, the same way.
If the Developer mode menu is not where the steps above say, ChatGPT renames these screens fairly often. Search Settings for "Developer mode," or ask a workspace admin if the account is on a managed plan.
What you get once it is connected
The same board, the same rules, whether the client asking is ChatGPT, Claude Code, or a plain REST call. The workflow is enforced by the server, not memorized by whichever model happens to be driving, so a task claimed from ChatGPT and a task claimed from another agent both go through the same lease and the same ranked list. Nothing about the coordination changes because of which client is attached.
Full setup, troubleshooting and FAQ live on the ChatGPT integration page; more engineering notes like this are on the blog.