Connect ChatGPT to an external memory: Developer mode, step by step

Updated 2 October 2026. Steps checked on chatgpt.com on 1 October 2026. ChatGPT's menus change often: if a label has moved, the principle stays the same.

To connect ChatGPT to an external memory, you go through Developer mode: on chatgpt.com, open Settings → Apps, turn on Developer mode under Advanced settings, create an app with the memory's MCP server URL, authorize it with OAuth, then enable it in the chat. It works on the web, on the Plus, Pro, Business, Enterprise and Edu plans, and the feature is still in beta.

This guide uses Pryzm Memory as the example: a persistent memory shared across several AIs over MCP. The steps are the same for any remote MCP server; only the name and the URL change.

What you need before you start

• A compatible ChatGPT plan: Plus, Pro, Business, Enterprise or Edu. The free plan does not include Developer mode.
• A web browser: apps are created on chatgpt.com. The mobile and desktop apps don't offer this setting.
• The MCP server URL of your memory. For Pryzm: https://pryzm-memory.com/mcp.
• An account with the memory service, for the authorization step. Pryzm is free up to 100 memories.
• On a Business or Enterprise plan, an admin may have turned Developer mode off for the workspace. If the option is missing, check that first.

The 5 steps in ChatGPT

1. Open settings. On chatgpt.com, open your profile menu, then Settings → Apps.
2. Turn on Developer mode. Under Advanced settings, switch on Developer mode. ChatGPT may show a warning about the risks of custom apps: read it, then confirm.
3. Create the app. Click Create app. Give it a name (for example “Pryzm”), paste the MCP server URL (https://pryzm-memory.com/mcp), choose OAuth authentication, then click Create.
4. Authorize access. A sign-in window from the memory service opens. Sign in to your account: ChatGPT receives its own access token, which you can revoke at any time from your account.
5. Enable the app in a chat. In a new chat, click “+” → Developer mode, then enable the app. ChatGPT can now call the server's tools: search the memory, save a memory, get a project overview.

To check that it works, ask a simple question that forces ChatGPT to use the app, for example: “Search my memory for what you know about my current projects.” ChatGPT should show the tool call before its answer. The first time, it may ask you to confirm the action.

Get ChatGPT to check the memory without asking every time

A connected app only helps if ChatGPT thinks of using it. By default, it calls the app when you ask explicitly. To make it check on its own, add an instruction to its custom instructions:

1. Open Settings → Personalization → Custom instructions.
2. Paste an instruction telling ChatGPT to check the memory at the start of every conversation and before answering anything that depends on your context, then to save what is worth keeping (decisions, preferences, tasks).
3. Save. The instruction applies to new conversations.

For Pryzm, the ready-to-paste text is on the connection page, in the “main memory” section. It fits within ChatGPT's character limit for custom instructions.

Limits worth knowing

• Web and paid plans only. No Developer mode on the free plan, nor in the mobile or desktop apps at the time of writing. A memory filled from the web can still be read by your other AIs.
• Beta feature. OpenAI presents Developer mode as a beta: its location, name and behaviour can change without notice.
• Re-enable per chat. Depending on the version, you may need to go through “+” → Developer mode in each new chat to enable the app.
• Write confirmations. ChatGPT may ask for your approval before each action that changes data, such as saving a memory. That's intended: you stay in control.
• You are responsible for the server you plug in. An app created in Developer mode is not an app from ChatGPT's catalogue: only connect MCP servers whose publisher you know.

What does the memory server see?

The server receives what ChatGPT sends when it calls a tool: a search query, or the text of a memory to save. It does not receive your full chat history, only what the tool call contains.

At Pryzm, servers are in Germany and each account is encrypted with its own key (AES-256-GCM). One limit worth knowing: during a session, the server decrypts your memories in memory so it can search them. It is not end-to-end encryption. A message that starts with ## is never saved.

If something doesn't work

• No Advanced settings or no Developer mode: check your plan, or ask your workspace admin (Business, Enterprise, Edu).
• Error when creating the app: check the URL, with no extra space or trailing slash, and that the authentication is set to OAuth.
• The sign-in window doesn't open: allow pop-ups for chatgpt.com, then try again.
• The app is connected but ChatGPT never uses it: check that it is enabled in the chat (“+” → Developer mode), then add the custom instruction.

When it isn't useful

If you only use ChatGPT, its built-in memory is probably enough. An external memory makes sense when you also work with Claude, Claude Code, Perplexity, Cursor or Gemini CLI, and want them to share the same context instead of re-explaining it to each one.

Guides for the other AIs are on the connection page. For Claude, see also how to add a custom connector.