How to share context between Claude, ChatGPT, and Cursor
If you use more than one AI in a day, you already know the pain. You explain your project to Claude in the morning, switch to Cursor at lunch to ship the code, then ask ChatGPT something at night and have to re-explain the whole thing. This is a practical guide to fixing that.
The problem: every AI is its own island
People who get the most out of AI today don't use one model. They use Claude for long-form reasoning, ChatGPT for quick lookups, Cursor for actually writing the code, Gemini when it's the right shape for the question. Each one is better at something than the others.
The cost is that none of them know what the others know. You ship a feature in Cursor. You try to debug it in Claude later. Claude has no idea what you built — so you paste the file, re-explain the architecture, re-state the constraints. Every switch is a tax on your time and a friction tax on your thinking.
It gets worse on bigger projects. Two weeks into a build, you have decisions, constraints, and naming conventions scattered across maybe four different chat histories in four different apps. None of them are searchable. None of them talk to each other.
Why the obvious fixes fall short
There are a few existing patches for this. They each cover part of the problem and miss the rest.
ChatGPT memory and Claude Projects
Both work, but only inside their own product. ChatGPT's memory is invisible to Claude, and Claude's Projects don't exist when you're in Cursor. The vendor incentive is to keep you locked in, not to share.
Copy-paste between chats
This is what most people actually do. It works, sort of, but it's slow, lossy, and you end up curating the same context block across five different threads. Two weeks in, you stop bothering.
Roll your own vector database
Pinecone, Weaviate, pgvector — all real options if you're a developer with time to wire each AI into them. Most people aren't building infrastructure. They want the answer to "what did I tell Claude last week" without writing code.
What changes with MCP
In late 2024, Anthropic published the Model Context Protocol — MCP for short. It's a standard for how AI assistants talk to external tools and data sources. Think of it like what HTTP is for the web, but for AI clients connecting to memory, file systems, calendars, and anything else they need.
The important thing for this guide: MCP is client-agnostic by design. Claude supports it. Cursor supports it. ChatGPT announced support. Gemini will get it via bridges. Anything that speaks MCP can read from and write to the same external memory.
That means a single memory server, plugged in once, can serve every AI you use. You no longer have to choose which AI gets to remember things — they all do.
How Neutrally fits
Neutrally is an MCP server that does one thing well: it remembers things for you, across every AI client you connect it to. Save a decision in Claude this morning; ask Cursor about it this afternoon and the answer is already there.
You don't bring your own API keys. You don't move between apps. You sign in to the AI you already pay for, and Neutrally sits behind it as a memory layer. Two MCP tools do most of the work:
- save_memory — the AI calls this when something worth keeping comes up (a decision, a preference, a constraint, a key detail about your project).
- search_memory— at the start of a new conversation, the AI calls this to pull in what you've told other AIs before.
The AI handles the actual calls. You don't type commands. You just talk to the AI like you normally would, and the memory layer threads itself into the conversation.
A concrete walkthrough
Here's what this looks like end to end. The full version takes about 5 minutes.
Free tier, 500 memories. No card required. You get a connector URL.
In Claude → Settings → Connectors → Add custom connector → paste the URL. You sign in via OAuth. Done. Claude can now read and write your memory.
Same URL, same flow. Each client connects to the same memory.
Tell it about your project. It will save the things that matter. You don’t see this happen — it’s a tool call under the hood. The memory accumulates.
Ask it about your project. Cursor calls search_memory, gets what you told Claude, and answers like it was there the whole time.
What it doesn't do
Worth being upfront on this. Neutrally is a memory layer — not a model, not a chat app, not a wrapper for ChatGPT or Claude. You still use whichever AI you prefer. We just give it memory it didn't have.
We also don't train on your data. Memories are stored under your account, deletable from the dashboard, and on the paid plan you can export everything as a portable file.
What it costs
Free tier covers 500 memories with every connector included. That's enough for most individual users. If you outgrow it, the Pro plan is £10 / month for unlimited memories, priority recall, and the export feature.
Connectors are never paywalled. We won't charge you to plug in a new AI client — every tier gets every supported client.
Why this matters beyond convenience
The shift here isn't cosmetic. When memory lives outside the model, the model becomes interchangeable. Today you might prefer Claude; in six months a better one will come out. Without portable memory, switching means re-onboarding the new AI from scratch. With it, you swap the model and keep everything you've built up.
That's what cross-AI memory really gets you. Not just less re-explaining today, but the freedom to use the best tool tomorrow without losing what you've already done.