If you use both ChatGPT and Claude, your knowledge is probably split across two sidebars, two “memories”, and a notes app neither of them can see. The durable fix is to make your notes app the library and let both assistants visit it. Thoughts does that with a single MCP connection: the same synced projects and thoughts are readable — and saveable — from ChatGPT and from Claude. This guide explains the setup, the habits that keep one library tidy, and why owning the shelf beats relying on either assistant's built-in memory.
The two-sidebar problem
Each assistant is helpful in the moment and forgetful across the wall. The meal plan Claude wrote is invisible to ChatGPT. The project summary in ChatGPT cannot be found from Claude. So you become the integration layer, copying context from one window to another and re-explaining decisions both tools already helped you make. Built-in memory features soften this inside one product, but they deepen the silo: everything the assistant “knows” about you lives in its house, in a format you cannot browse, file or take elsewhere.
A second brain, properly understood, was never meant to be a chatbot feature. It is your external library — notes you own, organised your way — that tools can consult. Assistants change; the library persists.
One library, two doors
Thoughts exposes one MCP server. Connect it in Claude using the Claude setup, connect it in ChatGPT using the ChatGPT setup, and both doors open into the same room: your synced projects and thoughts, with your titles, your filing, your collaborators. A recipe saved from a Claude conversation on your laptop is on your iPhone in the supermarket and searchable from ChatGPT next month when you want a variation.
Because the library is a real notes app, it also works with no AI involved. You capture by voice on a walk; on-device intelligence tidies and files the thought; later, whichever assistant you open can find it. The AI is a guest with a library card, not the librarian.
How a week actually flows
Monday, in Claude, you work through a launch plan and save the decisions to your Launch project. Wednesday, in ChatGPT, you ask what was decided about pricing; it searches Thoughts, reads Monday's thought in full, and drafts the customer email from it — which you save back to Launch. Friday, on your phone with no chat open at all, you voice-capture a stray idea; Thoughts files it under Launch. Sunday's weekly review can pull all of it together from either assistant, because there was only ever one pile.
Notice what did not happen: no exporting, no “as we discussed in the other tool”, no hunting through two histories. The project is the unit of continuity, which is exactly how organising notes by project was designed to work before AI entered the picture.
Rules that keep one brain from becoming one mess
File by project, never by assistant
Resist projects called “ChatGPT stuff”. A recipe belongs in Recipes whoever drafted it; that is where future-you (and future-assistant) will look. Authorship is tracked on each thought anyway.
Save conclusions, not conversations
Ask each assistant for the distilled, standalone version before saving. Two assistants saving raw transcripts would double the noise. The distil-and-save workflow in saving ChatGPT conversations applies verbatim to Claude.
One canonical version, edited in Thoughts
When a draft improves, update the thought in the app rather than saving a rival copy from the other assistant. Assistants create; you curate. That division is also the product's design: the connector can create thoughts but cannot edit or delete them.
Use the privacy split
Connected does not mean everything. Projects stay local unless you sync them, and only synced, unlocked content is reachable by either assistant. Keep the journal local; sync the cookbook and the project plans. One brain can still have a locked drawer.
What about each assistant's own memory?
Use it for preferences — tone, formats, how you like a list laid out. Do not use it as storage for artefacts you need to retrieve verbatim, share with a collaborator, or keep if you switch tools. Memory is a manner; your library is the record. If an assistant remembers that you prefer metric recipes, wonderful — but the recipe itself should live in Thoughts, where the other assistant can read it too.
Getting started
Connect the assistant you use most first and save three things this week. Then connect the second and ask it to find one of them. That small moment — the other tool just knowing — is when the two sidebars collapse into one brain. The connection guide keeps both setups and the starter questions in one place.
Frequently asked questions
Can ChatGPT and Claude really use the same notes library?
Yes. Thoughts provides one MCP server that both assistants can connect to. They read and save into the same synced projects and thoughts, so work saved from one is available in the other.
How is this different from ChatGPT or Claude memory?
Assistant memory stores preferences and facts inside that product, invisible to other tools and hard to browse or export. A Thoughts library is your own notes app: titled, filed, shareable and usable with no AI involved.
Will two assistants saving notes create a mess?
Not if you file by project rather than by assistant, save distilled conclusions instead of transcripts, and edit the canonical version in Thoughts. Assistants can create thoughts but cannot edit or delete, so curation stays with you.
Do both assistants see all my notes?
No. Both see only cloud-synced, unlocked thoughts in projects you own or belong to. Local-only projects and locked thoughts remain private on your device.