I wanted to sell a digital download for XMR without a payment processor, without accounts, and without anything on the clearnet. So I wrote a small shop and put it behind a Tor onion service. It went live this week. Some notes from the first day, because a few things surprised me.
The setup
- One Go binary and one SQLite file. No JavaScript, no cookies, no customer accounts.
- Every order gets its own subaddress. The buyer gets a 64-character receipt code instead of an account; the shop only stores its hash.
- The server has a view-only wallet. The spend key never touches it, so someone who takes the server still can’t move the money.
- No local node (“light mode”). Three copies of the view-only wallet each talk to a different remote .onion node, and a payment only counts once 2 of 3 report it identically with 10 confirmations. Any single honest node blocks a fake payment unless the other two lie the same way.
- The server is about 20 GB and uses about 3 GB.
Day one
- The first real order (a 0.001 XMR test) was recorded with 2 of 3 wallets agreeing, flipped to Paid, and the download worked.
- Over the first ~22 hours, the three wallets missed their 30-second poll 10, 47 and 118 times. My first instinct was “one node is bad, swap it.” But the misses clustered at the same moments across different nodes, and when I tested the “bad” node directly over Tor it answered 3/3 while another one failed 2/3. That pattern points at Tor latency on the server, not at any one node. Swapping a node would have just moved the problem.
- Because of the 2-of-3 rule, one slow wallet costs nothing. Two slow at once means a payment waits for the next cycle. Nothing gets recorded without agreement, which is the whole point.
- If I were starting again I’d probably run 5 nodes and require 3, which tolerates two slow ones at the same time.
Things it deliberately doesn’t do
No refunds in software, no subscriptions, no clearnet. Light mode trusts a majority of the nodes you pick; if that’s not good enough for what you sell, run your own node (~100 GB).
Disclosure: I’m selling the write-up
I turned the whole thing into a guide and bundled it with the complete source and a setup script that installs the stack on Debian/Ubuntu (it checks the Monero binaries against their signed hashes). It’s 0.052 XMR, sold from the same shop it teaches you to build.
The table of contents and one full section, on choosing remote nodes, are below for free, so you can judge whether it’s worth it.
Shop: http://2puqu7h2c6nr447zun3t6kdanhaq2xaq5xybnge7ivx4nrfidbtf4lqd.onion (Tor Browser)
Happy to answer questions about the setup here, whether or not you buy anything.
Free preview: Run Your Own Monero Onion Shop
Contents
- What you get, and who this is for
- How it works
- Two decisions before you start: own node or light mode, and where to host it
- Prepare your wallet (on your own computer)
- Prepare the server
- Build and upload
- Install
- Choosing remote nodes (light mode) ← full section below
- Addresses, product and launch
- Go-live checklist
- Running the shop
- Limits you should know about
- Troubleshooting
8. Choosing remote nodes (light mode)
-
Onion addresses only. The wallet talks to the node over plain HTTP. An onion address authenticates the node end to end; a clearnet node reached through Tor could be tampered with by the exit relay. The script refuses non-onion nodes.
-
Three different operators, ideally found on two different public node lists. The software cannot check that they are independent; you can.
-
Test each one from the server before using it:
curl -s --socks5-hostname 127.0.0.1:9050 -m 45 http://NODE.onion:PORT/get_info | jq '.height, .synchronized'Try a few times. Nodes reached through Tor often fail one request and answer the next.
-
Write down a spare in case one goes away.
What to expect in practice. On the reference shop’s first day, all three wallets timed out now and then, and the timeouts tended to happen at the same moments for different nodes. That points at Tor latency on the server rather than one bad node. One wallet missing a cycle is harmless: payments continue with the other two. When two miss the same cycle, payments wait for the next one, 30 seconds later. Nothing is ever recorded without 2 agreeing wallets.
You can use up to 5 nodes (wallet --light N1 N2 N3 N4 N5). A payment
then needs 3 of 5, which tolerates two slow nodes at once.
The full guide, the complete Go source with tests, and the install script: 0.052 XMR at http://2puqu7h2c6nr447zun3t6kdanhaq2xaq5xybnge7ivx4nrfidbtf4lqd.onion
Nice write-up, and the 2-of-3 view-only setup is clever. On the missed polls: I had the same pattern with several wallets over one Tor client. Giving each wallet its own circuit (a separate SocksPort, or a different SOCKS username with IsolateSOCKSAuth) and raising the RPC timeout to 60–90 s made most of the “bad node” misses disappear. Tor sometimes just has a slow minute for everything at once.
Also, once you’re happy with it, add the shop to a few Monero directories (XMRList and others list onion-only shops). That’s mostly how people looking to spend XMR find new places.
Thanks, glad the 2-of-3 setup makes sense to you.
Your point about Tor matches what the guide reports. On the reference shop’s first day, all three wallets timed out now and then, often at the same moments for different nodes. That pointed at Tor latency on the server rather than one bad node. Nothing is lost when that happens: nothing is recorded without 2 agreeing wallets, and a missed cycle just waits for the next one 30 seconds later.
The guide doesn’t currently cover giving each wallet its own circuit (separate SocksPort or IsolateSOCKSAuth) or raising the RPC timeout. I’ll check how that would fit the setup script and follow up.
Thanks also for the directory suggestion.
Glad it helped. Of the two, IsolateSOCKSAuth is the least invasive: you keep one Tor process and just give each wallet a different SOCKS username, so the setup script only changes the proxy line per wallet. Good luck with the launch.
Follow-up with numbers, as promised. I went with a separate SocksPort per wallet before I saw your IsolateSOCKSAuth note, and raised the RPC timeout from 15 s to 60 s. Over the ~34 hours since:
- failed payment checks (fewer than 2 of 3 wallets answering): 0.41/h → 0.12/h
- wallet timeouts overall: 9.5/h → 4.8/h
- wallet 3 alone: 4.25/h → 0.35/h
Wallet 2 barely improved, and that one turned out to be the node itself: it stopped answering entirely and has dropped off monero.fail. I’ve switched wallet 2 to another onion node, and all three wallets have answered every check since. So you were right about Tor, and my “it’s not one bad node” was only half right. Early numbers, but a clear improvement. Thanks for the tip, and for the directory pointer.
Thanks for coming back with numbers, that’s a clean result. Going from 0.41/h to 0.12/h on failed checks is a big drop.
Since one of the three nodes simply died, it might be worth automating the swap next time: a small script that calls
get_infoon each wallet’s node every few minutes, checks thatheightis within a block or two of the others, and moves the wallet to the next node on a short list if it falls behind or stops answering. monero.fail’s list works as a source for that fallback list, as long as you only keep onion nodes with a long uptime.Good luck with the shop.