Stop re-explaining yourself to ChatGPT, Claude and Cursor

Name: justonce.ai
Author: justonce.ai

Every AI vendor keeps memory in its own silo, so ChatGPT cannot see what you told Claude. JustOnce is a separate memory layer your assistants connect to over MCP. Save a fact once and Claude, ChatGPT, Cursor, Gemini and any other MCP client recall the same fact, in every later session.

Get started free Prompts to copy

Why doesn’t my context follow me when I switch AI tools?

Because memory is stored by the vendor, not by you. ChatGPT’s saved memory lives in your OpenAI account. Claude’s lives in your Anthropic account. Cursor keeps its own rules files. None of them can read the others, and none of them will — memory is the thing that keeps you from picking the best model for each task. So you re-explain the same projects, the same preferences and the same decisions, several times a day, or you maintain a context document you paste into every new chat.

JustOnce is not a feature inside one of those products. It is a separate memory service your assistants connect to at https://mcp.justonce.ai, so the memory belongs to your account rather than to whichever chat product you happened to open.

What actually happens once two assistants are connected?

You tell Claude on Monday that a client bills at a fixed monthly retainer and hates being called a “partner”. On Thursday you open ChatGPT to draft their newsletter, and it already knows both. Nothing was pasted, nothing was re-typed. The assistant asked JustOnce and got the same fact Claude wrote.

Connecting takes one command per tool — claude mcp add --transport http justonce https://mcp.justonce.ai — or a one-click deeplink in Cursor and VS Code, or a pasted connector URL in Claude Desktop and ChatGPT. Sign-in is OAuth; there is no API key to copy around. Recall and search are free on every plan, including Free.

What happens when a fact changes?

It supersedes rather than overwrites. Tell any connected assistant that the retainer went up in August and the old figure is not deleted — it is marked as what was true until August. That matters the first time you need to reconstruct what you charged in July, or work out which version of a decision a document was written against.

Memories are bitemporal. Corrections create a new value and keep the previous one, so your assistant can answer “what was true as of 1 July” as well as “what is true now”, and show every value a fact has held.

Can I see what it remembers about me?

Yes — that is the point of the design. Every memory is a visible record in the web app that you can read, correct, scope or remove. Each one keeps your exact words alongside the tidied-up version, so you can always check what was actually said against what was summarised. Recall results carry a confidence tier, so a shaky inference does not get presented to you as a fact.

Verbatim-first storage keeps the raw text on every memory. Confidence tiers run from VerifiedFact (0.90+) down, and are surfaced in recall results. Forgetting a memory is a soft, reversible invalidation rather than a hard delete.

What does JustOnce not do?

It does not read your chat history. Nothing is captured automatically from ChatGPT, Claude, Slack, a CRM or your mailbox — a memory exists because you or your assistant saved it, you uploaded a file, or you forwarded an email to your own ingest address. Recall is also scoped to your own vault: memories that someone else shared into a space with you are read in the web app, not returned by your assistant’s search. We would rather tell you that here than have you find out in week two.

Every recall tool resolves to the caller’s personal space. There is no space parameter on any of them, and there are no connectors that read a third-party product on your behalf.

Copy and paste

Prompts for this

Paste these into any assistant you've connected to JustOnce. They work as written.

Save a decision so every tool has it

Stores it as a durable memory, classified and with the people and organisations in it extracted. Any assistant you connect can recall it afterwards.

Remember this: we bill Acme a fixed monthly retainer, invoiced on the 1st, and they specifically do not want to be called a "partner" in any copy.

Pick up in a second tool where the first left off

Searches your vault and reads the standing context back, so a fresh chat in a different product starts where the last one finished.

Before we start, check JustOnce for what I already know about the Acme newsletter project and summarise it back to me.

Correct something without losing the old value

Supersedes the fact rather than overwriting it. The previous value stays available in the memory’s history.

Update what you know about the Acme retainer — it went up from 1 August. Keep the old amount in the history.

Ask what was true back then

Runs an as-of query and returns the value that was valid on that date, not the current one.

What was the Acme retainer as of 1 July this year?

Audit what it holds about a topic

Returns the memories with their confidence tiers, so you can spot anything stale or wrong and correct it.

List everything you have stored about Acme, and tell me how confident you are in each one.

Get set up in one line

Adds JustOnce to Claude Code. Run it once, sign in with OAuth, and the assistant can recall and save from then on.

claude mcp add --transport http justonce https://mcp.justonce.ai

The full prompt library

FAQ

Questions

Does JustOnce work with ChatGPT and Claude at the same time?

Yes. JustOnce is a remote MCP server, and each assistant connects to the same endpoint under your own account. Claude, Claude Code, ChatGPT, Cursor, VS Code, Windsurf and Gemini can all be connected at once, and they read and write the same memory. Free accounts get two connections; paid plans are unlimited.

Do I have to remember tool names or commands?

No. You talk to your assistant in plain language and it decides which JustOnce tool to call. “Remember that I bill Acme monthly” saves a memory; “what do I know about Acme” searches for one. The 24 tools exist for the assistant, not for you.

Is my memory used to train AI models?

No, never. Your memories are not used to train models by us or by anyone else. Data is hosted in Australia on Azure, memory bodies and documents are encrypted at rest, and sensitive details are vaulted separately and kept out of AI and search.

What happens if I stop paying?

Everything you have already saved stays searchable, because recall and search are free on every plan including Free. Paid plans meter what you add each month, not what you read back. Nothing is deleted when an allowance runs out.

How is this different from ChatGPT’s built-in memory?

Built-in memory is locked to one vendor, keeps only the latest value, and is hard to inspect. JustOnce serves the same memory to every connected assistant, keeps the full history so nothing is silently overwritten, and shows you every fact it holds so you can correct or remove it.

Can I bring what ChatGPT already knows about me?

Yes. Ask ChatGPT to list its saved memories, then paste them into a connected assistant and ask it to import them into JustOnce. Bulk import loads 50 at a time, and every new account gets a one-time welcome bonus of extra memories you can spend on exactly this.

Which AI tools can connect?

Any MCP-compatible client. Claude Code and the Gemini CLI connect with a single command, Cursor and VS Code have one-click deeplinks, and Claude Desktop, ChatGPT and Windsurf take a pasted connector URL. There is no browser extension.