If you use ChatGPT and Claude side by side, you know the problem: what you explained to one AI app yesterday, the other one doesn’t know today. Both have a built-in memory by now, but each lives in its own account and only knows its own conversations. This guide connects both to one shared memory. Afterwards Claude can read what you saved in ChatGPT, and the other way round. You need one address and about five minutes per app.

Why the built-in memories stay separate

For its memory, ChatGPT draws on past chats, saved memories and your custom instructions, and depending on your plan, on Library files and connected apps such as Gmail. All of that lives in your OpenAI account. Claude sees none of it. And if you set a ChatGPT project to “project-only memory”, you even split knowledge within ChatGPT (more on ChatGPT memory).

Claude’s memory summarizes your conversations on claude.ai, separately per project. You can import saved memories from ChatGPT into Claude once (how that works). That’s a move, not a connection: whatever ChatGPT learns afterwards never reaches Claude.

A shared memory therefore has to live outside both apps. The way there is called MCP, the Model Context Protocol: an open standard that lets AI apps use the tools of an outside service. ChatGPT and Claude both speak it.

What you need

  • A memory with an MCP server. This guide uses Memexa (from €2.99 a month, 14-day trial, card required). Other hosted MCP memories follow the same pattern.
  • Claude: any plan, the free one included. It allows exactly one custom connector; the paid plans allow several.
  • ChatGPT: we’ve tested the route ourselves on the Plus plan. We haven’t tested whether it works on the free plan.

The address is the same for both apps:

https://brain.memexa.ai/mcp

Step 1: connect Claude (claude.ai)

In Claude, MCP servers are called connectors. The route is the same in the browser and in Claude Desktop. If you’re signed in to claude.ai, this direct link opens the form straight away.

  1. Open Settings in Claude, then Connectors.
  2. Click Add custom connector.
  3. Name it “Memexa”, enter the URL https://brain.memexa.ai/mcp, confirm.
  4. Sign in with your Memexa account in the browser window. The sign-in stays valid.

More detail, including Claude Code, is in Set up Memexa in Claude.

Step 2: connect ChatGPT (chatgpt.com)

In ChatGPT, MCP servers are now called plugins; the same thing used to sit under “connectors”.

  1. Open chatgpt.com/plugins and click Add.
  2. Choose Add custom MCP server.
  3. Enter a name (“Memexa”), https://brain.memexa.ai/mcp as the server URL, and OAuth for authentication.
  4. Confirm the notice (“I understand …”) and click Create as plugin.
  5. In the browser window, sign in with the same Memexa account you used in Claude.

According to OpenAI, setup works in the browser only, not in the mobile app. OpenAI’s help page on custom MCP servers so far lists the feature only for Business, Enterprise and Edu, plus Pro with read permissions. The route works on Plus all the same; we’ve checked it ourselves in a Plus account (as of November 2026).

Step 3: check that both read the same memory

Tell Claude:

“Remember in Memexa: the quote for the Berger account goes out on Friday.”

Then ask ChatGPT:

“Check Memexa: when does the Berger quote go out?”

If ChatGPT answers “Friday”, both apps are reading the same memory. If it doesn’t work, check in both apps that the connector or plugin is signed in with the same Memexa account.

Important: nothing is saved automatically

This is where most expectations go wrong. MCP gives the AI app tools: search, read, create a note. When it uses which, the app decides during the conversation. It doesn’t record every chat and file it away on its own.

  • Reading usually works unprompted once a question sounds like it’s about your knowledge. Early on, “Check Memexa …” helps.
  • Saving happens when you say so: “Remember …”, “Save this decision to Memexa”, “Make a note of that”. Before changes, ChatGPT asks for confirmation depending on your settings. Under Settings → Plugins → Permissions you decide whether it always asks or, say, reads without asking and only asks before changes.
  • Make it a standing instruction: you can give the app this wish as a fixed instruction, in ChatGPT as a custom instruction, in Claude in a project’s instructions. For example: “When we decide something, save it to Memexa.” That makes saving more frequent, not complete.

Documents don’t reach the memory through the conversation at all: you upload PDFs, scans and mail exports to Memexa. They go through OCR, and both apps then cite the source down to the page.

What happens to ChatGPT’s and Claude’s own memory

Nothing. Both built-in memories keep running and don’t clash with the shared one. A split that works: preferences about writing style and answer length stay in each app, while facts, decisions and documents go into the shared memory. Whatever both apps should know belongs there.

Frequently asked questions

Does this work with ChatGPT Free?

We tested the route on the Plus plan. We haven’t tested the free plan. If chatgpt.com/plugins shows you the option for a custom MCP server, give it a try. Claude allows one custom connector on its free plan too.

Do OpenAI and Anthropic then see my whole memory?

No. The app only gets what a tool returns for a specific request, say three matching notes. Those results then become part of the conversation and are treated like anything you type yourself. The memory itself stays with Memexa in the EU (all details to check).

Does this also work with Mistral Vibe or Cursor?

Yes. Any AI app that speaks MCP reads the same memory, at the same address. The routes are listed under How it works: connect your AI apps.

What does it cost?

For Claude the free plan is enough; for ChatGPT we tested Plus. Memexa costs from €2.99 a month (Basic), Pro €9 and Power €19 (all plans). What the Claude plans cost is in Claude pricing in euros; what ChatGPT’s built-in memory can do and where it stops is on the ChatGPT memory page.