Korean Stock Data API (aikstockdata)

Why it's ranked here

The case rests on one field and on candour about it. The project attaches the hour and minute each filing was received, which it says the official DART search only gives as a date, and sorts filings into before-open, intraday and after-close. That split decides whether a filing could have moved that day's closing price. It also publishes market-adjusted returns after each filing type with 95% intervals, withholds numbers for types with fewer than 20 events, and states that 41 of 56 filled cells include zero. Setup is a single address with no credentials, where it says most Korean stock servers need a key first.

What's good

Every server reply carries an as-of date, a source link and a not-investment-advice line, and the instructions say a null means not provided, not zero. The server also tells the model that the tool list is not the whole catalogue, after a user's AI wrongly concluded earnings data was missing. The README warns that most AI fetch tools cut responses near 150 KB and points oversized files to smaller versions. The Python client has zero required dependencies, always sends a User-Agent because the default one gets a 403 from the CDN, and warns rather than silently padding a stock code given as an integer.

Tradeoffs

It is not real-time: prices are confirmed closes published once per trading day around 18:10 to 18:30 Korea time, and polling faster returns the same values. The licensing is split and not fully consistent. Code is MIT, but the server instructions say the data is non-commercial with attribution and commercial redistribution is not permitted, while the Python package docstring says it can be freely redistributed. The Python changelog marks its first version as not yet uploaded to PyPI, though the README shows a pip install. None of the files read contain the job that builds the published data. Most documentation is in Korean.

How to use it well

Best for someone asking an AI assistant about Korean listed companies after the close: what filed today, which filings landed after the market shut, how a filing type has historically moved prices. Check the catalogue's freshness status before quoting numbers, and when asking about post-filing returns, ask for the interval too, as the README itself advises. Treat stock codes as six-character strings. For direct file access, use the small variants of large files so AI fetch tools do not truncate them. It does not cover intraday quotes, consensus forecasts or commercial redistribution, so trading systems and resale products need another source.

Observed

License

MIT for the code; data terms stated separately as non-commercial with attribution

Languages

Python client library, JavaScript MCP server

Interface

Hosted remote MCP server over streamable HTTP, no authentication, read-only tools

Other access

Public JSON and CSV files over plain HTTPS; Python client

Python packaging

Requires Python 3.9 or later, no required dependencies, optional pandas extra

Container

Dockerfile runs a stdio-to-HTTP bridge to the hosted server rather than a local implementation

Data cadence

Once per trading day after the close; prices are T+1 confirmed closes, not real-time

Tests

Python client contract test runs as a standalone script, no test framework