Guides
Connect ChatGPT to Nawame: a step-by-step guide
Short answer
To connect ChatGPT to Nawame, turn on developer mode in ChatGPT settings, create an app for a remote MCP server with the address https://hub.nawame.eu/api/mcp, then enter the six-digit code sent to your e-mail. This follows OpenAI’s documentation as read on September 30, 2026: menus and availability vary by plan and workspace.

ChatGPT talks to Nawame through developer mode, which is the route OpenAI describes for plugging in a remote MCP server. This guide follows OpenAI’s official page on developer mode (read on September 30, 2026) together with Nawame’s own connection procedure.
One caution: ChatGPT’s menus change often. What you see may not match what follows, depending on your plan, your workspace or the version of the interface. If a label does not match, OpenAI’s page wins.
What you need
Three things are enough.
- An eligible ChatGPT account. OpenAI’s page lists Pro, Plus, Business, Enterprise and Education accounts, on the web. In a Business or Enterprise workspace, an administrator may have to allow developer mode first.
- A Nawame account, or the ability to create one. It is created on first access if you do not have one, and the free plan goes up to 10 memories (create an account).
- The server address, a single one, path included:
https://hub.nawame.eu/api/mcp
Nawame is a remote MCP server using streamable HTTP, with OAuth 2.1 authorization and dynamic client registration: no key to manage by hand, no software to install. ChatGPT calls it from the cloud, not from your computer, so the address has to be publicly reachable. It is.
Turn on developer mode in ChatGPT
In ChatGPT, open settings, go to “Security and login”, and switch on “Developer mode”.
OpenAI describes this mode as full MCP client support, with read and write tools. It also calls it powerful but risky, meant for people who know how to configure and test it, and names two dangers: prompt injection, and the model making mistakes on write actions. That warning explains the confirmations ChatGPT will ask for later.
If the option is missing, look at your plan and your administrator’s approval first.
Create the app with Nawame’s address
Open the ChatGPT Plugins page (chatgpt.com/plugins) and click the “+” button. That is where you create an app in developer mode for a remote MCP server. Exact field labels may vary, but here is what goes in:
- the server address:
https://hub.nawame.eu/api/mcp, copied in full, path included; - the protocol, if the form asks for one: OpenAI says ChatGPT supports SSE and streamable HTTP, and Nawame speaks streamable HTTP;
- the authentication: OAuth.
On credentials, OpenAI says static credentials, if provided, are used; otherwise ChatGPT may rely on Client ID Metadata Documents or on dynamic client registration (DCR), depending on the setup. Nawame can register a client dynamically, so you normally have nothing to type. If ChatGPT asks for a client ID and secret, see “If it does not work”.
Once the app exists, it shows up in the “Developer mode” tool of the composer (the + menu) and it is listed under “Drafts” in the app settings.
Authorize with the code sent by e-mail
At authorization time, Nawame sends a six-digit code to your e-mail. You type it in and you are done. If you did not have an account, it is created at that moment.
Nawame only returns to ChatGPT what your account is allowed to see: the AI never exceeds your rights. The exact screen depends on the client. For the general principle, the documentation describes the Claude side of things.
Check that it works
Open a new conversation and switch on your app from the “Developer mode” tool. OpenAI advises being explicit about which tool you want. One detail first: the trigger phrases Nawame documents are French (« active Nawame », « n’enregistre plus », « qu’as-tu retenu ? », « ouvre NW-XXXXXX »). We have no official English wording, so the English versions below are our own equivalents. Try them, then check what the AI actually did. A few simple requests, in order:
- “Which Nawame tools can you see?” The model can be wrong when it describes itself: a hint, not proof.
- “Activate Nawame.” In French, « active Nawame » opens a memory and starts saving as you go.
- A real request. Imagine someone redoing their association’s website: “We keep two menu levels at most, because most visitors are on mobile. Record that as a decision.” ChatGPT should ask you to confirm before it writes anything.
- “What did you keep?” (« qu’as-tu retenu ? ») The AI rereads what it has saved since the start of the exchange.
- In a fresh conversation, “Open NW-XXXXXX” (with the topic’s anchor code, shaped like NW-2ACRFY) picks the topic up where you left it.
To stop saving, say so (« n’enregistre plus » in French). Saving ends, and what is already written stays.
In a one-off chat with no lasting project behind it, Nawame does nothing: if nothing happens for a pancake recipe, that is not a fault.
Good habits with write actions
According to OpenAI, write actions ask for confirmation by default. You can remember your choice, for the current conversation only. Tools not flagged as read-only are treated as writes, so even something that looks like a plain lookup may still ask for your approval.
A few habits that pay off:
- read what ChatGPT is about to write before you confirm, and check the JSON payloads of the calls, as OpenAI suggests;
- the app settings let you turn tools on or off, and refresh the app to reload the server’s tools and instructions;
- Nawame treats stored content as data, not as instructions, but that does not replace your own attention in a mode OpenAI itself calls risky.
The “skill” file is an extra, not another way to connect: Nawame offers a SKILL.md file to upload (in ChatGPT: Plugins, Skills, “+”, Upload, if the labels have not moved). Adding it does not connect Nawame, you still need the connector.
If it does not work
- The menus do not match. An older label seen in the product help was “Settings, Apps & Connectors, Developer mode, Create”. Menus evolve: follow OpenAI’s page.
- Developer mode is missing. Check your plan, web access (the page mentions the web) and your administrator’s approval.
- Nawame’s tools do not show up. Refresh the app from its settings. If that fails, create a new one with the same address. Some chat clients show a shorter tool list than code-oriented clients.
- ChatGPT asks for a client ID and secret. Use Nawame’s fixed OAuth client: in the Nawame console, My account, Advanced section, “Fixed OAuth client”. Paste the redirect URL the client requires, create it, copy the client ID and the secret (the secret is shown only once), then enter them in the app.
And if you just want to stop: disconnecting Nawame from an AI client deletes neither the account nor the memory.
Current limits
- A mode its own publisher calls risky. OpenAI insists on caution, especially for writing.
- Variable availability. Plan, workspace, web access, labels: all of it can change, and a chat’s tool list may be shorter than a code-oriented client’s.
- A procedure checked on one date. The steps above follow OpenAI’s documentation as read on September 30, 2026, plus Nawame’s procedure, nothing more. These features change fast.
- Sovereign mode. Stronger privacy can reduce what the server can process, and therefore some features (see the security page).
Where to go next
Once the connection works, try it on a real topic that lasts: using Nawame on a long-running project walks through an example over several weeks. If you switch AI tools along the way, how to move project context from one AI to another compares the methods, including the plain manual brief when that is enough.
Another AI? See Mistral and Grok, each with its own labels. For more general questions, the FAQ covers the most common ones.


