6 months, 46K queries, 10 power users, $0 marketing spend
Shibui Finance is an MCP server that gives Claude access to 64 years of US stock market data. Users connect it and ask questions in plain English. No code, no subscriptions.
I posted "The Start" here in May. Here is what happened since.
The numbers
- 46,000+ SQL queries generated and executed
- 96% success rate (Claude writes the SQL, not the user)
- 10 confirmed returning power users
- ~800 tool calls/day
- ~69 conversations/day
- Server cost: one VPS. Total infrastructure spend is under $50/month.
What users actually built
I expected people to use it for basic stock screening. Instead, 10 returning users built completely different analytical systems on top of it.
One built a 25-year tax-aware backtesting engine through conversation. Another built a daily insider-cluster detection system with anti-gaming filters. Someone else built a compliance-based screener layering religious finance rules on top of standard financial filters. A fourth built a walk-forward factor validation system and uses a second AI model to adversarially review the results.
None of these were use cases I designed for. Every user arrived with an existing methodology and discovered they could finally execute it at scale without writing code.
What worked for distribution
SEO. I publish landing pages on shibui.finance targeting long-tail queries like "consecutive earnings growth screener" and "best MCP server stock data." ~200 daily search impressions, 8-17 clicks/day. This is the primary acquisition channel and the only one that compounds.
Bing Copilot. Shibui gets cited 12-16 times per day in Bing Copilot AI answers. I didn't do anything to make this happen. It started appearing and keeps growing month over month.
Articles. Published on dev.to, HackerNoon, and Hashnode. Each gives a dofollow backlink. Different angle per platform so nothing is duplicated.
What didn't work
Twitter. ~120 tweets over 4 months. Zero conversions. Not a single user came from X. I stopped posting in June.
Substack outreach. Emailed ~40 finance Substack writers offering free access. Got polite replies, zero sustained usage. Closed the channel in March.
RIA outreach. 291 emails to registered investment advisors. ~50% click-through rate on early batches, but converting clicks to active users is a different problem. Still running at low volume.
What I learned
The product sells itself to people who already know what they want to ask. They arrive with a methodology, try it, and come back. The problem is finding those people, not convincing them.
Retention is fine (41-65% of direct-connect users return). Acquisition is the bottleneck. SEO is slow but it compounds. Everything else I tried was a one-time spike that decayed to zero.
If I were starting over, I would skip Twitter entirely and put that time into landing pages from day one.
What's next
Authentication and a free tier cap. Right now it is completely free with no login required. That needs to change before I can measure anything properly.
A weekly newsletter bridging the landing page to the MCP server. Two drafts sitting in a folder, not shipped yet.
More landing pages. The SEO channel is working. I just need to keep feeding it.
Try it: shibui.finance
The difference between the use cases you expected and what those 10 power users actually built is pretty striking. Curious which one surprised you the most.
I expected users screening for stocks more or less casually, but turns out the users are real professionals and run multi factor screens and backtests. That was quite a surprise because when I started building this I had no idea such things exists.
I've been learning so much by observing what users do with the service.
That’s a meaningful signal — the users are revealing a much more sophisticated job than you originally expected. If you’re open to continuing the conversation, what’s the best email to reach you at?
This is a great breakdown of the gap between measurement assumptions and real user behavior. You expected basic stock screening but users built 25-year tax-aware backtesting engines, insider-cluster detection systems, compliance screeners. That's the classic blind spot - your measurement system (what you thought users needed) was completely wrong, but usage patterns revealed what they actually valued.
The other insight that stands out: SEO compounds, everything else decayed to zero. Twitter had zero conversions despite 120 tweets. Substack outreach got polite replies, zero retention. But search traffic keeps growing. That's the difference between building on owned channels vs rented attention - distribution channels determine which feedback signals actually reach you.
Yup, I've somewhere heard "Inbound is king", and now I'm following this advice, the compounding effect is the best.