Moving from ChatGPT to Claude?

Bring your projects, not just a memory summary.

ChatGPT's official export doesn't include your memories, custom instructions, or project structure. Claude's import tool copies a summary of your memories and preferences — once, into Claude only. The OneLore kit moves the knowledge itself into portable documents that Claude, Le Chat, Cursor, and every MCP-capable ai can read and keep building on.

What actually transfers

	ChatGPT's export ZIP	Claude's memory import	OneLore kit
Saved memories		summary, experimental	verbatim, structured
Custom instructions		summary, experimental	verbatim, portable
Project briefs & decisions			one brief per project
Full chat history	raw JSON		on our roadmap
Where it lands	a ZIP on your disk	inside Claude only	portable docs any ai reads

Sources: OpenAI's export documentation and Anthropic's import-memory tool, as of July 2026. Both are good at what they do — this kit covers what they don't.

Before you start

You need a free OneLore account — sign in with Google on your first connection, no separate signup. Then pick a path: direct (ChatGPT writes to OneLore itself) or copy-paste (works on every ChatGPT plan, including Free).

Path A — directChatGPT Plus · Pro · Business · Enterprise · Edu

ChatGPT's Developer Mode can connect to OneLore and write your knowledge over directly. Free and Go plans don't expose Developer Mode — use Path B instead.

1
Connect OneLore to ChatGPT: Settings → Apps → Advanced → enable Developer Mode, then Create app with the server URL https://mcp.onelore.ai. Sign in with Google when prompted. Full walkthrough on /get-started.

In a regular chat, run the migration prompt. ChatGPT asks you to confirm each write — approve them; there are at most four.

migration prompt

You have access to my OneLore tools (lore_*). Migrate my ChatGPT knowledge into my personal OneLore project.

1. Find where to write. My personal project is named "Me". If it is my only OneLore project, leave projectId empty on every write — it resolves automatically. If I have several projects, list them first and use the projectId (the id, not the word "Me") of the one named "Me".

2. Write out everything you know about me from your saved memories and my custom instructions — verbatim where possible, grouped by topic. Do not summarize away details. If you cannot reproduce something word-for-word, paraphrase it as closely as you can and keep every specific (names, tools, formats, dates); do not drop an item just because you can't quote it exactly.

3. Organize it into at most four documents:
   - docs/chatgpt-import/profile.md — who I am: role, background, people and tools I work with
   - docs/chatgpt-import/preferences.md — how I like to work: style, formats, dos and don'ts
   - docs/chatgpt-import/projects.md — every project or ongoing effort you know about, with current state and key decisions
   - docs/chatgpt-import/notes.md — anything important that fits nowhere else
   Keep each document under 50,000 characters. If a topic is larger, split it into numbered parts (projects-1.md, projects-2.md) rather than exceeding the limit — that matters more than staying at four files.

4. Write each document with lore_context_write. Start every file with this YAML frontmatter (fill in today's date):

---
created: YYYY-MM-DD
created_by: chatgpt-migration
description: <one line describing this file>
---

5. When you are done, list what you wrote and anything you chose to leave out.

Write full prose, not fragments — these documents will be read by other ai assistants and by me.

For each ChatGPT Project you care about, open a chat inside that projectand run the brief prompt. Project files and instructions are only visible from within the project — a regular chat can't see them.

project brief prompt

Using this project's files, instructions, and our conversation history in this project: write a complete project brief — purpose, current state, key decisions made (and why), open questions, and preferences specific to this project.

Then save it with lore_context_write at docs/chatgpt-import/projects/<short-project-name>.md. Write it to my personal OneLore project named "Me": if it is my only project, leave projectId empty; if I have several, pass the projectId (not the word "Me") of the one named "Me". Start with the standard YAML frontmatter (created: today's date, created_by: chatgpt-migration, description: one line). Keep it under 50,000 characters.

Path B — copy-pasteevery ChatGPT plan, including Free

No connector needed on the ChatGPT side. ChatGPT writes the export as text; Claude (with OneLore connected) files it.

In ChatGPT, run the dump prompt and copy the output. For projects, run it again inside each ChatGPT Project and copy those too.

dump prompt (run in ChatGPT)

Write out everything you know about me — your saved memories and my custom instructions — verbatim where possible, grouped under these headings:

## Profile
## Preferences
## Projects
## Notes

Do not summarize away details; include specifics (names, tools, formats, dates). If you cannot reproduce a memory word-for-word, paraphrase it as closely as you can rather than dropping it. Output one single markdown block I can copy.

2
Connect OneLore to Claude — one paste on /get-started, sign in with Google.

In Claude, run the filing prompt with your export pasted underneath.

filing prompt (run in Claude)

Below is an export of my ChatGPT memories and custom instructions. File it into my personal OneLore project using lore_context_write.

Where to write: my personal project is named "Me". If it is my only OneLore project, leave projectId empty; if I have several, pass the projectId (the id, not the word "Me") of the one named "Me".

Split it into up to four documents under docs/chatgpt-import/ (profile.md, preferences.md, projects.md, notes.md). Start each file with YAML frontmatter (created: today's date, created_by: chatgpt-migration, description: one line). Keep each file under 50,000 characters — split into numbered parts if larger. Keep the content faithful — organize it, don't rewrite it. When you are done, list what you wrote.

[paste your ChatGPT export below this line]

Check it worked

Ask any connected ai:

Read docs/chatgpt-import/ in my OneLore project (the one named "Me") and tell me what you now know about me.

From here your context follows you: connect Claude Desktop, Claude Code, Le Chat, Cursor, or any MCP-capable ai and every one of them reads the same documents. They are plain Markdown in your own project storage — no lock-in, EU-hosted, never used for training.

What this doesn't migrate

Full conversation history. ChatGPT's official export (Settings → Data controls → Export) gives you the raw JSON; a deep importer that turns it into structured context is on our roadmap.
Uploaded files and attachments. Re-upload the ones that matter to your new home.
Custom GPTs. There is no export for them — but pasting a GPT's instructions into either path's prompt preserves the knowledge.

Connect your first ai

Free tier fits a full migration comfortably — the kit writes at most a handful of documents.