Export chats and messages
Use an assistant that can save files on your computer: Claude Code, Cursor, Codex, or the Claude app with file creation turned on. Not connected yet? Start with “Connect in five minutes”.
The data lands in two files you can reuse: one row per chat, and one line per message.
Scope it first. Knowing how many chats you're about to pull tells you whether to take everything or a sample.
A file is the difference between a one-off answer and an analysis. Once the chats are on disk, every chart, comparison and re-run works from the same data, and you don't have to read the whole inbox again for each question.
- 1
Check the size before you pull.
PromptHow many chats had at least one message in the last 30 days, split by WhatsApp number? Don't export anything yet. - 2
Export the chat list.
PromptExport every chat active in the last 30 days to chats.csv with these columns: chat ID, WhatsApp number, contact name, labels, assigned teammate, status, created date, last message date and message count. Shorten phone numbers to the last 4 digits.chats.csv, exported from the demo workspace through MCP - 3
Export the messages behind them.
PromptFor the chats in chats.csv, save every message to messages.jsonl, one line per message: chat ID, time, direction (customer or us), sender and text. Work in batches of 50 chats and tell me the progress. - 4
Check that nothing is missing.
PromptCheck the export: does chats.csv have the same number of chats you counted, and does every chat have at least one message in messages.jsonl? List any gaps.
This is customer data The files hold names and real conversations. Keep them on your own computer, leave out what you don't need, and delete them when the analysis is done. Big inbox? Start with one number, one week or a random sample of 200 chats.