ClaudeMarkdownBackupPrivacyProductivity

How to Back Up Claude Chats to Markdown, Then Delete Them Safely

Zephyr Whimsy2026-11-177 min read

My Claude sidebar had gotten out of hand. Hundreds of conversations: half-finished debugging sessions, drafts, and research threads I would never open again. A few of them had answers I really wanted to keep. I wanted to save the good ones as Markdown files I could search, put under version control, or paste back into an AI later, and then delete the rest without worrying that I had lost something.

This post describes the workflow I ended up using. I tested it on my own account over about a week. It covers two methods, the official bulk export and a per-chat conversion in the browser, plus the checks I run before deleting anything.

Why Markdown and not just "export"

Claude's official export is useful, but it gives you JSON and not files you would want to read. Markdown is better for three reasons:

  • It is plain text. You can grep it, diff it, and open it in Obsidian, VS Code, or any notes app.
  • It keeps structure. Headings, lists, and especially code blocks survive with their formatting intact.
  • It goes straight back into AI tools. When I continue a project, I paste the old conversation into a new Claude chat or a Cursor context window, and Markdown is the format these tools handle best.

Method 1: The official Claude data export (bulk, JSON)

If you want everything, start here. In Claude, open Settings, then Privacy, and request a data export. You get an email with a download link to a zip file. Inside is a conversations.json file with every chat.

What I liked:

  • It is complete. Every conversation, every message, timestamps included.
  • It is first-party, so nothing depends on a third-party tool.

What I did not like:

  • It is one large JSON blob. My file was several megabytes, and it is not readable without a script.
  • Artifacts, attachments, and some formatting do not always come through the way they looked in the UI.
  • It is all or nothing. You cannot export only a few chats.

If you are comfortable with a short script, you can loop over the JSON and write one .md file per conversation. I did this as a safety net. I treat it as a cold archive that I will probably never read, not as my working library.

Method 2: Convert individual chats to Markdown in the browser

For the conversations I actually wanted to reuse, which was maybe 40 out of several hundred, I converted each one directly from the open chat page.

The catch is that Claude conversations sit behind your login. Server-side converters such as Jina Reader or Firecrawl are good tools for public pages: they are fast, scriptable, and great in pipelines. But they fetch the URL from their own servers, so they get your login screen and not your chat. MarkDownload is a solid open-source extension that runs in the browser, and it works fine on many pages. On chat interfaces, though, I got mixed results with how messages and code blocks were split.

I used Web2MD for this step because it runs inside the page you are already logged into. The conversion happens locally in the browser, so the conversation text is not sent to a third-party server to be scraped. That mattered to me, because some of these chats contained internal code and notes from clients.

My process for each chat:

  1. Open the conversation in Claude and scroll to the top so the whole thread is loaded.
  2. Click the Web2MD icon to convert the page.
  3. Check the token counter. It tells me right away whether this chat will fit in a new context window if I paste it back later.
  4. Save it as a .md file with a dated, descriptive filename.

This is what a converted chat looked like for me:

# Debugging Postgres connection pool exhaustion

**User:**
Our Node service starts throwing "too many clients" after about 2 hours.
Pool size is 20, max_connections is 100. What am I missing?

**Claude:**
The most common cause is connections that are checked out but never
released, usually in an error path. Check for code like this:

    const client = await pool.connect();
    const res = await client.query(sql); // throws
    client.release(); // never reached

Wrap the query in try/finally so release() always runs...

The User and Claude turns are clearly separated and the code survives as code. I can drop this file into a repo's /docs folder, and it reads like a normal note.

For research-heavy chats, I add a short header by hand before saving, so future me knows why it exists:

---
source: claude.ai
saved: 2026-11-12
project: pricing-page-rewrite
keep_because: final copy decisions + rejected options
tokens: ~6,800
---

# Pricing page copy review

**User:**
Here are three headline options for the Pro tier...

The tokens line comes straight from Web2MD's counter. It sounds like a minor detail, but it is how I decide whether to paste a whole file into a new chat or summarize it first.

The delete step: what I check first

Deleting from Claude is permanent from your side. There is no trash folder. Before deleting anything, I go through this checklist:

  1. Open the saved .md file and skim it. Make sure it is not truncated. Long chats sometimes load lazily, so if you did not scroll up, the top part can be missing.
  2. Check that code blocks closed properly. A missing closing fence will break the rendering of everything after it.
  3. Check for artifacts. If Claude produced a separate artifact, such as a document or a component, copy that content out on its own. Do not assume it was captured inline.
  4. Keep the official export zip somewhere safe as a backstop, even if you never plan to open it.

After that, I delete. You can delete single chats from the conversation menu. For bulk cleanup, the chat list view lets you select several conversations and delete them together, though the exact UI changes from time to time. I deleted in batches of about 20 and spot-checked my Markdown folder between batches.

Honest limits

A few things to know before you rely on this:

  • Web2MD is Chrome-only for now, which includes Chromium browsers like Edge and Brave. Firefox and Safari users need a different tool.
  • The free tier allows 3 conversions per day. That was enough for me to trickle through my archive over a couple of weeks. If you have 200 chats to save this afternoon, you will hit the limit. Pro is $9/month and removes the cap (see pricing).
  • It is per-page, not bulk. For a full archive, the official JSON export is still the right tool. I use both: JSON for completeness, Markdown for the chats I will actually reuse.
  • Very long chats can be large. One of my threads came out at over 40k tokens. The counter flagged it, and I split it into two files.

My final setup

After a week of cleanup, here is what I have:

  • One zip of the official export in cold storage.
  • A claude-archive/ folder with about 40 Markdown files, named like 2026-11-12-pricing-copy-review.md, all searchable with ripgrep and synced to my notes app.
  • A Claude sidebar with only active projects in it.

The best part is reuse. When I return to an old problem, I open the .md file, check the token count, and paste it into a fresh chat as context. That often works better than scrolling through an old thread, because I can trim the parts that went nowhere before pasting.

If you want to try the per-chat method, install Web2MD and convert one conversation you care about. The free tier covers that. Read the output, and if it holds up, you will know whether the workflow suits you before you delete anything. More guides on getting web content into AI tools are on the Web2MD blog.

Related Articles

Most Read

last 30 days
  1. #1Can Claude Read Reddit? No — Here's the One-Click Fix (2026)
  2. #2HTML vs Markdown for LLMs: I Wasted 67% of My Tokens for a Year
  3. #3Reducing Token Waste in ChatGPT and Claude: 7 Techniques That Cut Costs 72%
  4. #4Obsidian Web Clipper Official Plugin 2026: Complete Guide + When You Need More
  5. #5Reddit JSON API vs Scraping: The Honest 2026 Comparison for Developers

Latest Articles