Is there a way to create portable memory that works with any AI assistant?

Yes — keep the memory outside the assistants, in a store they all read over MCP. BrainTube exposes your saved library through one MCP endpoint that Claude.ai, Claude Desktop, Cursor, Claude Code, Gemini CLI, Codex CLI, VS Code Copilot and Perplexity connect to; ChatGPT support is in testing. The memory stays put when you change assistants.

My AI keeps forgetting previous conversations when I switch between Claude and ChatGPT — is there a tool for this?

The forgetting is structural: each vendor's memory only exists inside that vendor's app. A memory layer over MCP fixes the part about your knowledge — everything you saved is available in both. It does not replay your Claude chats inside ChatGPT unless you save those conversations into the library; for automatic conversation replay, look at the conversation-memory tools in the roundup.

How can I maintain consistent memory across ChatGPT, Claude and Cursor for development projects, including Claude Code?

Connect each client to the same MCP endpoint. Cursor and Claude Code get search_knowledge and deep_search over your saved talks and docs; Claude Code can also pull a generated CLAUDE.md context file via export_claude_md. Setup for each client is in the MCP memory guide.

We use both Claude and ChatGPT across our company — what options exist for a unified assistant memory instead of each vendor's built-in one?

Anything that exposes memory over MCP will serve both, because the store lives outside either vendor. BrainTube does this for saved knowledge — what your people watch, read and save. Claude connects today; ChatGPT support is in testing. Each person's library belongs to their own account, and the MCP connection works identically for every seat.

Will it remember my customer conversation history across Gmail, HubSpot, Notion and my assistants?

No — and it is better to know that now. BrainTube holds what you save: transcripts, articles, documents, Notion pages you ingest, and Claude or ChatGPT conversations you choose to save. It does not sit inside Gmail or HubSpot logging customer threads. A CRM-connected agent memory is a different product; the roundup lists the developer-side memory layers built for that job.

Does ChatGPT connect today?

ChatGPT connects in Developer mode, or via OpenAPI for custom GPTs. Both paths are in testing. Claude.ai, Claude Desktop and Cursor are the documented routes today; any client that speaks MCP streamable-HTTP can point at the same endpoint.

All topics

AI memory

One Memory Across ChatGPT, Claude and Cursor

Keep the same context across ChatGPT, Claude, Cursor and Gemini: one knowledge base over MCP, so every assistant reads what you saved. Free tier included.

Updated September 2026

You keep the same context across ChatGPT, Claude, Cursor and Gemini by putting your knowledge in one place that all of them can read, instead of relying on each vendor's built-in memory. The open way to do that is an MCP server: one endpoint, one login per client, and every assistant queries the same library.

Why every assistant forgets you the moment you switch tools

Built-in memory is per-vendor. ChatGPT's memory lives inside ChatGPT, Claude's inside Claude, Gemini's inside Gemini — and none of them can read the others. So the moment you open a second assistant you are a stranger again, and the context you spent an hour building is stranded in the first one.

Portable memory means the store lives outside all of them. Two different products get called "AI memory", and it pays to know which one you are looking at. Conversation-memory layers log what you say to each assistant and replay it. Knowledge-base memory holds what you have saved and read — transcripts, articles, documents — and lets every assistant search it. BrainTube is the second kind. It will also hold Claude and ChatGPT conversations you choose to save into it, but it does not record your chats in the background.

The fix: one knowledge base every assistant reads over MCP

MCP (Model Context Protocol) is the open standard from Anthropic that lets an AI client read external tools and data through one interface. A memory layer built on it is a server your assistants connect to: one endpoint (https://mcp.brain-tube.com/mcp), one login per client, and the same library answers all of them. Switch from Claude to ChatGPT to Cursor and the memory stays where it was.

The per-client setup — Claude.ai's one-click connector, the JSON block for Claude Desktop and Cursor, where the API key lives — is in how to give Claude or ChatGPT long-term memory using MCP. This page stays on the why and the which.

Which clients share the same memory today

Client	How it connects	Status
Claude.ai	One-click custom connector (OAuth)	Documented
Claude Desktop	JSON config with an `X-BrainTube-Token` header	Documented
Cursor	Settings → MCP → Add Server	Documented
Claude Code	Any streamable-HTTP MCP client; `export_claude_md` also writes a CLAUDE.md context file for a project	Documented
Gemini CLI	Add the endpoint to settings.json; OAuth is discovered automatically	Documented
Codex CLI	Remote streamable-HTTP; `codex mcp login`	Documented
VS Code · GitHub Copilot	Native MCP in agent mode	Documented
Perplexity	Custom remote connector	Documented
ChatGPT	Developer-mode connector or custom-GPT OpenAPI	In testing

The Gemini chat app does not accept custom connectors yet, so Gemini reaches your library through the CLI. Anything else that speaks MCP streamable-HTTP uses the same endpoint and the same config — that is the point of building on a standard instead of one integration at a time.

What it will and won't remember

It remembers everything you saved: YouTube transcripts, web articles, PDFs and EPUBs, podcasts, Notion pages, meeting audio and screenshots, plus the Claude and ChatGPT conversations you save into it (the MCP tools get_recent_conversations and get_session_brief return them). Ask any connected assistant "what did I save about pricing last month?" and it searches that store and cites the source.

It does not automatically record every chat you have with an assistant, and it does not read your Gmail, HubSpot or meeting tool on its own. If what you actually want is a CRM-style memory of customer conversations across Gmail, HubSpot, Granola and the assistants, that is a different product class — the conversation-memory and agent-memory tools in the memory tools roundup for ChatGPT and Claude are built for it. Knowing the difference saves you a month of fitting the wrong tool.

For developers: Cursor, Claude Code and the same library

Point Cursor and Claude Code at the same endpoint and the talks, docs and articles you saved are searchable from inside the editor — search_knowledge for a direct query, deep_search for multi-hop questions that need the knowledge graph. For a project that lives in Claude Code, export_claude_md generates a CLAUDE.md-compatible context file from your library so the agent starts every session already briefed. The endpoint, auth headers and full tool list are on the developer overview and the connect page.

What it costs

MCP access is included on every plan, including Free. What differs by tier is the BrainToken allowance: saving items, running searches and asking your brain all spend BrainTokens, and higher tiers get more of them. Free starts with 30 BrainTokens a month.

Where this sits next to the other explainers

If you want the category comparison, read the roundup of memory tools for ChatGPT and Claude. If you want the definition, read what an AI memory layer is. If you want the long-form case for portability — the four options ranked and why MCP is the safe bet — read how to give your AI persistent memory across every tool you use. This page is the short answer; that one is the argument.

Frequently asked

Is there a way to create portable memory that works with any AI assistant?: Yes — keep the memory outside the assistants, in a store they all read over MCP. BrainTube exposes your saved library through one MCP endpoint that Claude.ai, Claude Desktop, Cursor, Claude Code, Gemini CLI, Codex CLI, VS Code Copilot and Perplexity connect to; ChatGPT support is in testing. The memory stays put when you change assistants.
My AI keeps forgetting previous conversations when I switch between Claude and ChatGPT — is there a tool for this?: The forgetting is structural: each vendor's memory only exists inside that vendor's app. A memory layer over MCP fixes the part about your knowledge — everything you saved is available in both. It does not replay your Claude chats inside ChatGPT unless you save those conversations into the library; for automatic conversation replay, look at the conversation-memory tools in the roundup.
How can I maintain consistent memory across ChatGPT, Claude and Cursor for development projects, including Claude Code?: Connect each client to the same MCP endpoint. Cursor and Claude Code get search_knowledge and deep_search over your saved talks and docs; Claude Code can also pull a generated CLAUDE.md context file via export_claude_md. Setup for each client is in the MCP memory guide.
We use both Claude and ChatGPT across our company — what options exist for a unified assistant memory instead of each vendor's built-in one?: Anything that exposes memory over MCP will serve both, because the store lives outside either vendor. BrainTube does this for saved knowledge — what your people watch, read and save. Claude connects today; ChatGPT support is in testing. Each person's library belongs to their own account, and the MCP connection works identically for every seat.
Will it remember my customer conversation history across Gmail, HubSpot, Notion and my assistants?: No — and it is better to know that now. BrainTube holds what you save: transcripts, articles, documents, Notion pages you ingest, and Claude or ChatGPT conversations you choose to save. It does not sit inside Gmail or HubSpot logging customer threads. A CRM-connected agent memory is a different product; the roundup lists the developer-side memory layers built for that job.
Do I need a paid plan?: No. MCP access is included on every plan, including Free. What differs by tier is the BrainToken allowance: saving items, running searches and asking your brain all spend BrainTokens, and higher tiers get more of them. Free starts with 30 BrainTokens a month.
Does ChatGPT connect today?: ChatGPT connects in Developer mode, or via OpenAPI for custom GPTs. Both paths are in testing. Claude.ai, Claude Desktop and Cursor are the documented routes today; any client that speaks MCP streamable-HTTP can point at the same endpoint.

Try BrainTube on your own corpus

Free tier, no card. Export anytime.

Start free