The problem: Claude and ChatGPT don't talk to each other
If you use both Claude and ChatGPT — one for drafting, one for research, or just because different teammates prefer different tools — you've hit the same wall: whatever you tell one of them stays with that one. Paste your project brief into Claude in the morning, and ChatGPT has no idea it exists when you open it after lunch. A decision your team wrote down in a ChatGPT thread is invisible the next time someone opens Claude to act on it.
This isn't a bug in either product. Claude and ChatGPT are built by different companies, with separate accounts, separate memory (where they have any at all), and no shared channel between them. The workaround most people fall into — copy-pasting context back and forth, or keeping a running doc you paste into both — works, but it's manual, it drifts, and it's easy to forget a step.
The fix is to give both clients the same external memory to read from and write to, so switching between them doesn't mean switching context.
How it works
Great Arrow Digital runs an MCP (Model Context Protocol) server that both Claude and ChatGPT can connect to independently, using the connection method each vendor supports. Claude connects as a custom connector inside Claude.ai's own connector settings; ChatGPT connects as a custom connector added through its developer mode. Both connections point at the same MCP endpoint (https://www.greatarrowdigital.com/api/mcp) and the same authenticated workspace — so once connected, they aren't two separate memories, they're two windows onto one memory.
Both connectors authenticate with OAuth 2.1 and dynamic client registration, which is what lets each client walk you through an in-app consent screen instead of asking you to copy and paste an API token. Whatever gets stored — a memory, a document, a synced integration fact — is encrypted at rest and becomes searchable to whichever of the two clients asks for it next.
Setting it up
- Create your account. Sign up at /sign-up. No credit card is required to start the 14-day trial.
- Open the install page and start with Claude. Go to /install, signed in, and choose Claude.ai (web). This client uses the guided OAuth path: in Claude's own settings, go to Settings → Connectors → Add custom connector, and complete the consent screen the install page walks you through.
- Add ChatGPT the same way. Back on /install, choose ChatGPT. It's also a guided OAuth connector: in ChatGPT, enable developer mode, then add a custom connector from Settings → Connectors → Add custom connector, and complete OAuth consent. ChatGPT's custom connectors require a Plus, Team, or Enterprise plan.
- Confirm both connections. The install page can detect a successful connection automatically for most clients. If detection doesn't fire for one, a manual "I did this" confirmation step is available so you're not stuck waiting.
- Test it. Tell Claude something specific — a project detail, a preference, a decision. Then open ChatGPT and ask a related question. If the connectors are set up correctly, ChatGPT's answer reflects what you told Claude, without you repeating it.
What each client sees after connecting
Once both are connected, there's no asymmetry: Claude and ChatGPT are peers reading and writing the same workspace. Anything one client stores — directly, or through a synced integration like Gmail or Slack — is available to the other. Neither client gets a "richer" or "stale" copy; both search the same underlying memory at query time, so the answer you get depends on what's actually stored, not on which client happened to store it.
What doesn't carry over is each vendor's own product-level memory feature, if you use one — Claude's or ChatGPT's built-in memory (where either offers it) stays local to that vendor's account and isn't part of this shared layer. The Great Arrow Digital memory is a separate, explicit layer both clients read from in addition to whatever their own vendor gives them.
Conversations themselves become shared context
The shared layer isn't limited to facts you deliberately stored. Transcripts from connected AI clients feed the same workspace memory — cross-client conversation persistence — so a discussion you had with Claude can inform ChatGPT's later answer even when nobody stopped mid-conversation to write the conclusion down. Important decisions and facts get auto-memorized from that stream, which is what closes the gap between "we talked about this somewhere" and a searchable fact either client can retrieve on demand.
That matters for a two-client setup specifically, because context tends to accumulate in whichever client you happened to be using at the time. Without transcript persistence, memory sharing works only as well as your discipline about explicitly saving things; with it, the baseline is that both clients share the trail of what's actually been discussed, and explicit saves become the exception for things you want pinned precisely rather than the only way anything survives.
Where to go next
If you're rolling this out to more than two clients — adding Cursor, Gemini, or a CLI tool into the mix — One Memory Across Every AI Client covers the general pattern and the full client list. If Gmail is part of what you want Claude searching, Claude MCP Server for Gmail covers connecting that integration specifically.