ChatGPT to Claude

Migrate all your ChatGPT conversations into Claude with two browser-console scripts — no API keys, no extensions, no external tools.

Note: Conversations are exported as a flat zip of .txt files — ChatGPT project/folder structure is not preserved. The importer places every conversation into a single Claude Project of your choosing.

Step	Script	Where to run	What it does
1	ChatGPTConversationExporter.js	chatgpt.com	Exports every conversation as a .txt inside a .zip.
2	ClaudeConversationImporter.js	claude.ai	Reads the .zip, lets you pick a Claude Project, and imports each conversation as a new Claude chat.

Both scripts work the same way: open F12 → Console, paste the script, hit Enter. They reuse the Authorization token and Cookie the browser already holds for the active session — nothing to configure.

---

Table of Contents

• Requirements
• Step 1 — Export from ChatGPT
  • Usage
  • How It Works
  • ChatGPT API Endpoints Used
  • Configuration
  • Conversation Format
  • Duplicate Filename Handling
  • Built-in Zip Writer
• Step 2 — Import into Claude
  • Usage
  • How It Works
  • Claude API Endpoints Used
  • Configuration
  • Built-in Zip Reader
• Privacy & Security
• UI Overlays
• Limitations & Known Constraints
• Troubleshooting
• License

---

Requirements

• A modern browser (Chrome, Edge, Firefox, etc.).
• An active, logged-in session on chatgpt.com (Step 1) and claude.ai (Step 2).
• No extensions, API keys, or external tools required.

---

Step 1 — Export from ChatGPT

Usage

1. Go to chatgpt.com and make sure you are logged in.
2. Press F12 (or right-click → Inspect) to open Developer Tools.
3. Click the Console tab.
4. Copy the entire contents of ChatGPTConversationExporter.js.
5. Paste it into the console and press Enter.
6. A full-screen progress overlay will appear — keep the tab open until the export finishes.
7. When complete, a chatgpt_export_YYYY-MM-DD.zip file will download automatically.

How It Works

1. URL Check — Verifies the script is running on chatgpt.com. If the hostname does not match, an alert is shown and the script exits immediately.
2. Authentication — Retrieves your session access token by calling /api/auth/session. The token is sent as a Bearer token in the Authorization header for all subsequent requests.
3. Discovery — Paginates through /backend-api/conversations (100 conversations per page, ordered by updated) to build a complete list of every conversation in your account.
4. Project Scan — Fetches all ChatGPT Projects via /backend-api/projects and, for each project, paginates through /backend-api/conversations?project_id={id}. Any conversations not already seen are merged into the main list. This catches conversations hidden behind the sidebar’s “Show more” ellipsis.
5. Fetch — For each conversation, fetches the full message tree from /backend-api/conversation/{id}.
6. Extract — Walks the message tree using a depth-first traversal (iterative stack). Each node’s message.content.parts array is joined into plain text. Messages are labelled by role (You, ChatGPT, System, Tool).
7. Package — Writes each conversation as a .txt file inside a zip archive using a built-in pure-JavaScript zip writer (STORE method, no compression). All files are stored flat (no folder nesting).
8. Download — Generates the zip blob in-browser and triggers an automatic download.

ChatGPT API Endpoints Used

Endpoint	Method	Purpose
/api/auth/session	GET	Retrieve the current session access token
/backend-api/conversations?offset={n}&limit=100&order=updated	GET	Paginate through the conversation list
/backend-api/projects?offset={n}&limit=100&order=most_recent	GET	Discover all ChatGPT Projects
/backend-api/conversations?project_id={id}&offset={n}&limit=100	GET	Paginate conversations inside a specific project
/backend-api/conversation/{id}	GET	Fetch the full message tree for a single conversation

Configuration

Variable	Default	Description
DELAY_MS	1200	Milliseconds to wait between API calls (rate-limiting / politeness delay).
BASE	/backend-api	Base path for ChatGPT backend API calls.

Conversation Format

Each .txt file is formatted as:

======================================================================
  Conversation Title
======================================================================
  Date: 1/15/2025, 3:42:00 PM

--- You ---
User message text

--- ChatGPT ---
Assistant response text

======================================================================
  End: Conversation Title
======================================================================

Roles are mapped as follows:

API role	Display label
user	You
assistant	ChatGPT
system	System
tool	Tool
anything else	raw role string

Duplicate Filename Handling

If two conversations have the same title, subsequent files get a numeric suffix:

Some Title.txt
Some Title (1).txt
Some Title (2).txt

Filenames are sanitized: characters < > : " / \ | ? * are replaced with _, leading/trailing dots are stripped, and names are truncated to 100 characters.

Built-in Zip Writer

The exporter includes a minimal, dependency-free zip writer implemented in pure JavaScript. It builds a valid ZIP archive using the STORE method (no compression):

1. Encodes file content as UTF-8 and computes a CRC-32 checksum for each entry.
2. Writes Local File Headers followed by raw file data.
3. Writes a Central Directory with one entry per file.
4. Writes an End of Central Directory (EOCD) record.
5. Returns the result as a Blob for immediate download.

Note: No external libraries are loaded. This avoids Content Security Policy (CSP) issues on chatgpt.com that block third-party CDN scripts.

---

Step 2 — Import into Claude

Usage

1. Go to claude.ai and make sure you are logged in.
2. Press F12 (or right-click → Inspect) to open Developer Tools.
3. Click the Console tab.
4. Copy the entire contents of ClaudeConversationImporter.js.
5. Paste it into the console and press Enter.
6. An overlay will appear. Click Minimize & Capture, then send any message in Claude's chat. This lets the script capture your session's auth headers.
7. The overlay reappears with a file picker — select your chatgpt_export_YYYY-MM-DD.zip file.
8. If you belong to multiple Claude organizations (e.g., Free + Team), a picker appears — choose your org.
9. A project picker appears — choose the Claude Project to import into. (Create one in Claude first if you haven't already.)
10. The script imports every conversation as a new chat inside the chosen project.
11. When finished, a summary screen shows conversations imported and errors. Click Close to dismiss.

How It Works

1. URL Check — Verifies the script is running on claude.ai. If the hostname does not match, an alert is shown and the script exits immediately.
2. Auth Capture — The overlay minimizes and the script intercepts window.fetch to capture auth headers and the active model name from the next Claude API call the user triggers (e.g., sending a message). These headers are replayed via un-patched native fetch (obtained from a hidden iframe) for all subsequent API calls.
3. File Selection — Presents a styled file picker (<input type="file" accept=".zip">).
4. Parse — Reads the zip entirely in-browser using a built-in pure-JavaScript zip reader (no external libraries). All .txt files are extracted into a flat list regardless of any folder structure in the zip. See Built-in Zip Reader.
5. Organization Picker — Fetches all Claude organizations from /api/organizations. If you belong to multiple orgs (e.g., personal Free plan + a Team workspace), a picker lets you choose which one. Single-org accounts skip this step.
6. Project Picker — Lists your existing Claude Projects and presents a picker. All conversations will be imported into the chosen project.
7. Create Conversations — For each .txt file, a new chat conversation is created via POST /api/organizations/{orgId}/chat_conversations with the conversation title and project_uuid. The conversation content is then sent as the first message via the streaming completion endpoint, and the response is drained so the server commits the conversation.
8. Summary — Displays a completion screen with conversations imported and errors.

Why does this work? Claude's frontend wraps window.fetch and adds auth headers (session tokens, client identifiers) internally. Our script captures those headers by intercepting a real API call, then uses un-patched native fetch (from a hidden iframe) to make its own requests with the same credentials. This bypasses the wrapper entirely while preserving full authentication.

Claude API Endpoints Used

Endpoint	Method	Purpose
/api/organizations	GET	Retrieve the current user's organization UUID
/api/organizations/{orgId}/projects	GET	List existing projects (for project picker)
/api/organizations/{orgId}/chat_conversations	POST	Create a new chat conversation (uuid, name, project_uuid)
/api/organizations/{orgId}/chat_conversations/{convId}/completion	POST	Send a message to a conversation (streaming response)

Configuration

Variable	Default	Description
DELAY_MS	800	Milliseconds to wait between API calls (rate-limiting / politeness delay).

Built-in Zip Reader

The importer includes a minimal, dependency-free zip reader implemented in pure JavaScript. It works by:

1. Scanning backwards from the end of the ArrayBuffer to locate the End of Central Directory (EOCD) signature (0x06054b50).
2. Reading the Central Directory entries to discover each file's name, compression method, compressed size, and local header offset.
3. Parsing each Local File Header to calculate the exact data start position.
4. Decompressing each entry: STORE (method 0) entries are used as-is; DEFLATE (method 8) entries are decompressed using the browser's built-in DecompressionStream API.
5. Decoding the resulting bytes as UTF-8 text.

Note: The exporter generates STORE-method (uncompressed) zips for maximum compatibility. The importer also handles DEFLATE-compressed zips produced by other tools. Directories and zero-length entries are skipped.

---

Privacy & Security

• No data leaves your browser — both scripts run entirely within your active browser session. Conversations are read from one service and written to another using the same browser that already has access.
• No third-party servers — neither script loads external resources. Both operate entirely using built-in browser APIs.
• No API keys — both scripts piggyback on the authentication tokens/cookies your browser already holds for chatgpt.com and claude.ai.
• Credentials are never stored — access tokens are used in-memory for the duration of the script and discarded when the page is closed.

---

UI Overlays

Both scripts render a full-screen overlay with a dark theme (rgba(0,0,0,0.85) backdrop) to show real-time progress:

Exporter Overlay (ChatGPTConversationExporter.js)

• Header — "ChatGPT Conversation Exporter" with a purple-to-green gradient.
• Progress bar — Animated fill from 0% to 100% as conversations are processed.
• Counter — {current} / {total} conversations.
• Status line — Current operation description.
• Current item — Monospace display of the conversation title being exported.
• Log panel — Scrollable panel showing warnings (yellow) and errors (red).
• Close button — Appears after export completes.

Importer Overlay (ClaudeConversationImporter.js)

• Header — "Claude Conversation Importer" with a warm brown/copper gradient.
• File selection view — A styled button to trigger the file picker.
• Organization picker — Buttons for each Claude org (shown only for multi-org accounts).
• Project picker — Buttons for each existing Claude Project.
• Progress view — Same layout as the exporter (counter, progress bar, status, log).
• Completion view — Checkmark icon, summary statistics, and a Close button.

---

Limitations & Known Constraints

Area	Detail
Rate limiting	Both scripts use a fixed delay between API calls (DELAY_MS). If you hit rate limits, increase the value.
Project structure	ChatGPT project/folder organization is not preserved. All conversations are exported as flat .txt files and imported into a single Claude Project.
Conversation size	Very large conversations may approach Claude's knowledge document size limits.
Zip compression	The exporter generates STORE (uncompressed) zips. The importer supports both STORE and DEFLATE. Other compression methods are not supported.
Browser tab	The export/import tab must remain open and in the foreground for the full duration of the operation.
Session expiry	Long-running exports (thousands of conversations) may encounter session token expiration. Re-run the script if this happens.
Message types	Only text content is extracted. Images, file attachments, and DALL-E generations are not included.

---

Troubleshooting

Problem	Solution
"…must be run on chatgpt.com / claude.ai"	You pasted the script on the wrong website. Navigate to the correct domain first. The exporter only runs on chatgpt.com; the importer only runs on claude.ai.
"No accessToken in session response"	Your ChatGPT session may have expired. Refresh the page, ensure you are logged in, and try again.
"Not a valid zip file (no EOCD found)"	The selected file is not a valid zip archive. Make sure you are selecting the .zip output from Step 1.
"Failed to fetch orgs"	Your Claude session may have expired. Refresh claude.ai, log in, and re-run the importer.
HTTP 429 errors	You are being rate-limited. Increase DELAY_MS (e.g., to 2000 or higher) and try again.
Script does nothing	Check the browser console for errors. Some browsers require you to type allow pasting first before pasting scripts into the console.
Missing conversations	The exporter skips conversations with zero extractable text messages. Check the log panel for "No messages" warnings.

---

License

This project is provided as-is for personal use. Use at your own risk.