Shibui Finance is built and maintained by Christian Richter, a data engineer based in Berlin with 15+ years of experience building ETL pipelines on cloud infrastructure. It is a bootstrapped project, not a venture-backed company. The same person designs the data pipeline, writes the MCP server, and answers the support email.
Who builds Shibui Finance?
Before Shibui, Christian spent over a decade designing and operating data pipelines for clients across industries - from financial fraud detection (ID Analytics) to cloud deployment platforms (AltusInsight, which he founded and ran for eight years). Earlier work included speech recognition research at Fraunhofer and a first-employee role at a Bay Area startup. He holds a Diplom in Advanced Electromagnetics from TU Ilmenau.
Shibui grew out of that pipeline experience applied to a specific problem: financial data that AI assistants could actually query, rather than hallucinate from training weights. The ETL architecture, the dbt transformation layer, the DuckDB export pipeline - these are patterns from 20+ data engineering projects, adapted for a domain where getting the numbers wrong has real consequences.
How is Shibui Finance built?
The whole stack runs on open Python and data tooling. A daily ETL job pulls end-of-day prices, fundamentals, valuations, and SEC filing metadata from licensed and public sources, loads them into PostgreSQL, and transforms them with dbt into clean, queryable tables. The MCP server exposes that database to Claude and ChatGPT through a single read-only query tool. There is no proprietary data feed and no black box: the data sources page lists every provider, the coverage window, and the refresh cadence.
Why does Shibui Finance exist?
Most financial tools either keep their methods opaque or hand you a dropdown that only sees today's value. Shibui was built around one idea: let people ask a real question in plain English and get the actual numbers back, with the query Claude ran shown alongside the answer. You can always check the work. The guides and examples show what that looks like in practice.
What are the data limitations?
The harder commitment is being honest about the data itself. Shibui does not run on Bloomberg or Refinitiv. Coverage is US equities only, end-of-day, split-adjusted but not dividend-adjusted, and the price history is survivor-biased because delisted tickers drop out over time. The site states all of this plainly, on the data page and in every guide, because a screening tool that oversells its inputs is worse than no tool. Shibui returns data, never financial advice.
Timeline
Shibui launched in early 2026. The ETL pipeline, database schema, and MCP server were built from scratch over several months. Since launch the database has grown to 31M+ daily price records, 56 technical indicators, and 6.4M SEC filing metadata entries covering 9,900+ US securities. The data pipeline runs daily on a VPS, with SEC filing feeds updating every five minutes.
Get in touch
Questions about coverage, a correction, or a feature request reach the person who built it: chris@shibui.finance. You can follow development on the blog, X, GitHub, and LinkedIn.