How to share memory between ChatGPT and Claude.
ChatGPT and Claude each keep their own memory, and neither reads the other's. But both speak the same connector protocol — so if you point them at one shared memory server, a fact you tell Claude is there when you ask ChatGPT. This is the general recipe, the exact steps for both sides, and what to expect, with Genesys as the worked example at the end.
Two assistants, two silos.
ChatGPT's memory and Claude's memory are each scoped to their own product. They do not sync with each other, and there is no setting that points one at the other. Tell Claude your stack, your timezone, the way you like answers formatted — none of it is there when you switch to ChatGPT for the next task. The context you built up is trapped in whichever app you happened to build it in.
The fix is not to copy memory back and forth. It is to move the memory out from under either assistant and into a neutral store that both connect to — so there is one memory, and the assistants are just two windows onto it.
One MCP memory server, connected on both sides.
This works with any assistant that supports MCP connectors and any compliant MCP memory server. The pattern, not the vendor, is what matters.
The Model Context Protocol (MCP) is an open standard for connecting assistants to external tools and data. A memory server is just an MCP server whose tools are store, recall, and friends, backed by a persistent store. Because the protocol is shared, the same server can be a connector in Claude and a connector in ChatGPT at the same time. Whatever one assistant writes, the other can read — the store is the single source of truth.
https://genesys-api.astrixlabs.ai/mcpConnect it in Claude.
The one-click link opens Claude's add-connector modal with the name and URL already filled in; you only approve and authenticate. Nothing is sent to the server until you sign in and grant access.
Connect it in ChatGPT.
ChatGPT does not have a prefilled connector link the way Claude does, so the button copies the MCP URL and opens the plugins page for you to paste it. Developer-mode connectors are a newer, evolving feature — expect to enable the connector once per conversation until that changes.
After both are connected.
Genesys as the shared store.
Everything above works with any MCP memory server. Genesys is the one we run: https://genesys-api.astrixlabs.ai/mcpis the endpoint you connect on both sides. It is open source, and the memory is yours — a single causal graph you can inspect, correct, export, or delete, not a black box inside someone else's app. Recall is scored, stale memories fade on evidence, and every recall can explain why it surfaced.
On the retrieval quality that makes cross-assistant recall actually useful, Genesys scores 85.55 ± 0.37 on LoCoMo, the mean of ten runs under a frozen gpt-4o-mini answerer and judge at temperature 0 (n=1,540, categories 1–4; evaluated July 2026) — published alongside its ceiling and the answers it loses as well as gains, not as a bare marketing number.
Common questions.
Can ChatGPT and Claude share the same memory?
Yes. Neither assistant reads the other's built-in memory, but both support Model Context Protocol (MCP) connectors. If you point both at the same remote MCP memory server, that server becomes shared memory: a fact you store in Claude is recalled in ChatGPT, and vice versa. The assistants stay separate; the memory underneath them is one store.
How do I share memory between ChatGPT and Claude?
Run or subscribe to an MCP memory server, then connect it on both sides. In Claude, add it under Settings → Connectors → Add custom connector (or use a one-click connector link). In ChatGPT, enable developer mode, then add the same server's /mcp URL as a custom connector and enable it in the conversation. Both assistants now read and write the same memory graph.
Do ChatGPT's and Claude's own memory features already do this?
No. ChatGPT memory and Claude memory are each scoped to their own app; they do not sync with each other, and you cannot point one at the other. Sharing memory across both requires a neutral store that both can connect to — which is exactly what an MCP memory server is.
Is this specific to Genesys?
No. The mechanism works with any compliant MCP memory server, self-hosted or hosted. Genesys is the worked example on this page because it is the memory server we run, it is open source, and its memory is user-owned and inspectable — but the steps are the same for any server that exposes an MCP endpoint and the memory tools.
What should I expect after connecting both?
Store a fact in one assistant and it is available to the other on the next turn that recalls memory. In Claude the connector persists once approved; in ChatGPT, developer-mode connectors currently must be re-enabled per conversation. Authentication happens once per client over OAuth, and recall adds a short retrieval round-trip to the request.
Related.
MCP memory servers explained
What MCP is, and how connectors and OAuth make a memory server portable across apps.
What is an AI memory layer?
Store, recall, forgetting; vector vs. temporal-graph vs. causal-graph architectures.
LoCoMo results
The retrieval-quality number behind useful recall, with the caveats stated.
Genesys overview
Connect once — every app that speaks MCP reads the same memory.