Since Claude Code came out I've been chewing on one question: now that writing a parser costs basically nothing, and an LLM can do a lot of the data validation that used to need a team — can you disrupt a Data-as-a-Service company as a solo dev? So I tried to build the thing and find out. US, Japan, and Korea (Europe not yet) all have standardized filing formats and XBRL taxonomies, so I wrote a pipeline that pulls structured financials straight out of the raw reports — income statement, balance sheet, cash flow, segments, ownership, the usual. This kind of data goes for anywhere from $99 to $20k/yr online depending on depth and accuracy. Short answer: it's doable. I bought a bunch of commercial datasets out of pocket to check myself against, and: - JP is the best — most standardized, and quality is basically on par with JQuants - US and KR are accurate enough for an individual researcher. Every value is backed by the raw XBRL fact so you can trace it. It's exposed as an MCP server, so it drops straight into Claude Code, Codex, etc. — you ask for a ticker's financials in plain language and it pulls from the parsed DB instead of scraping. Seed users are free right now; I mostly want feedback on where the data is wrong. Would love to hear what breaks, and which markets/fields people actually want.

Rockmoon

Financial data you can audit

Financial data you can audit