Did Claude Code leak your email?
Drop in your Claude Code transcripts and see every web request that had an email address in its User-Agent header, with the site and the date. Your files stay in this tab.
Your transcripts
Drop your .claude/projects folder here
Windows: %USERPROFILE%\.claude\projects. Mac and Linux: ~/.claude/projects. On a Mac, press Cmd+Shift+. in the picker to see hidden folders. Your browser may ask to “upload” the folder. That is its standard wording: the files are read here, in this tab, and nothing is sent.
What happened
Claude Code can see the email address of the signed-in account. Some sites ask automated clients to identify themselves, and users have reported the model filling that in with their real address when it writes a request. The reports include curl calls to crates.io, a subagent calling liquipedia.net and web searches.
I ran this checker on my own machine: 1,206 transcript files, 1.9 GB. It found 4 User-Agent headers with an address in them, on 2 sites. One was a script calling reddit.com.
How the checker works
A transcript records every tool call the agent made. The checker reads only those calls, because they are what actually ran. Your address also sits in the session’s context, and that alone isn’t a leak, so context and tool output are ignored.
- User-Agent headers. The address must sit inside the header’s value. This is the headline count.
- URLs and contact fields. Addresses in a query string or an email field of a request are listed separately, because many are deliberate.
- Did it run? Each call is matched to its result, so you can tell a request that went out from one that was refused.
Addresses are hidden until you choose to show them, so you can share your screen. The copied result has no addresses or site names in it.
If it finds something
- Update Claude Code. A comment on the open issue says version 2.1.317 and later include a built-in mod,
cc-plugin-privacy, that blocks this and is on by default. Check it under/plugins. - Don’t rely on that alone. Later comments in the same issue report gaps, including scripts the model writes itself.
- Add a line to your
CLAUDE.mdtelling it to use a generic User-Agent and never to put your address in a request. - Run the checker again after your next few sessions.
Frequently asked
Where does Claude Code store its transcripts?
In a .claude/projects folder inside your home folder, with one .jsonl file per session and a subfolder for each project. On Windows that is %USERPROFILE%\.claude\projects. On a Mac or Linux it is ~/.claude/projects.
Are my transcripts uploaded?
No. The page reads the files in your tab with JavaScript and has no server code that could receive them. You can check: open devtools, watch the Network tab, and run a scan. Or load the page, turn off your internet, and run it then. The site counts page views like any other, but nothing from your files goes into that.
Does a result mean a site has my email?
It means the address was written into a request and the tool call ran. It doesn't prove the site kept it, and a site that answered with an error may still have logged the request. The checker can't see that side.
Why does a hit say Blocked or failed?
The transcript records the result of each tool call. If the call was refused, for example at a permission prompt, the checker marks it that way, because the request did not go out.
What could it miss?
It only finds an address written out in full in the tool call. It can't see an address read from a variable or a config file at run time, or one sitting in a script that was saved earlier and run later. It also skips placeholder addresses, such as anything at example.com. So zero hits means none were found, not that none were ever sent.
Why are some hits under worth a look?
An address in a URL or a form field is often deliberate, such as signing in to your own app or looking up a customer. Those stay out of the headline count, so you can read them and decide. Requests to localhost and git commit details are skipped.
Does it work with other AI coding tools?
Not yet. It reads the transcript format Claude Code writes. Files from other tools will scan without errors and find nothing, so don't read that as a clean result.
Built by Himanshu Srivastava, a backend engineer in Bengaluru. For more on your own Claude history, see Claude Unwrapped. To see how the scan works, read how to check Claude Code transcripts for your email. To check your settings files for saved keys and passwords, try the Claude Code settings secret scanner. To see what the same transcripts would cost at API prices, use the Claude Code usage cost calculator. For another check that keeps your data in the tab, try the JWT decoder, or see all the tools.