If you model Asian economies, build a macro dashboard, or give an AI assistant a "what's inflation in Korea?" tool,
you eventually need South Korea's official numbers. They live in two places:
- KOSIS — Statistics Korea's portal (CPI, employment, population, births…)
- ECOS — the Bank of Korea's database (base rate, exchange rates, money supply, GDP…)
Both have free official APIs. In practice two things slow people down: you register for a key on each portal, and
KOSIS returns table, item and category names in Korean, so you end up building your own translation and parsing
layer before you can plot anything.
This post shows a shortcut: two small Apify actors that call those official APIs for you and return clean,
English-labelled time series — no Korean account or API key on your side.
What you get back
One row per observation, with English labels, a numeric value, the unit and the source:
| table | period | item | value | unit |
|---|---|---|---|---|
| KOSIS DT_1J22003 | 2026-10 | CPI, whole country | 120.73 | 2020=100 |
| KOSIS DT_1DA7001S | 2026-10 | Unemployment rate, total | 2.0 | % |
| ECOS 722Y001 | 2026-10 | Bank of Korea base rate | 3.0 | % per annum |
| ECOS 731Y001 | 2026-11-13 | KRW per USD | 1332.4 | Won |
A small sanity check I like: CPI comes from Statistics Korea in KOSIS and is mirrored by the Bank of Korea in ECOS.
Pulled independently, both returned 120.05 / 120.43 / 120.73 for Aug–Oct 2026.
Every row also carries an attribution block (source, table, the official table URL, retrieval date), so a chart
or an AI answer can cite where the number came from.
1. Python: several KOSIS tables in one call
# pip install apify-client pandas
from apify_client import ApifyClient
import pandas as pd
client = ApifyClient("<YOUR_APIFY_TOKEN>")
run = client.actor("kdata-hub/kosis-korea-stats").call(run_input={
# ORGID/TBLID from the kosis.kr table URL; up to 20 tables per run
"tables": ["101/DT_1J22003", "101/DT_1DA7001S"],
"periodType": "M",
"startPeriod": "202401",
"endPeriod": "202610",
})
df = pd.DataFrame(client.dataset(run["defaultDatasetId"]).iterate_items())
cpi = df[(df.table_id == "DT_1J22003") & (df.item_id == "T") & (df.category_code == "T10")]
unemp = df[(df.table_id == "DT_1DA7001S") & (df.item_id == "T80") & (df.category_code == "0")]
print(cpi[["period", "value"]].tail(3))
print(unemp[["period", "value"]].tail(3))
Note on size: this pulls every item and category of both tables — 1,634 rows for this period (about $3.27).
To fetch only what you need, run one table at a time with "itemId" and "classL1" set (CPI total:
"itemId": "T", "classL1": "T10"; unemployment rate total: "itemId": "T80", "classL1": "0") — 34 rows each.
Any table you can open on kosis.kr works — the address bar shows orgId=…&tblId=…. You don't need to know how many
classification levels a table has; the actor detects that. If one table id is wrong, the other tables still return
and the bad one is named in the run log (a failed table produces no rows, so it isn't charged).
2. Python: Bank of Korea rates and FX
run = client.actor("kdata-hub/ecos-korea-stats").call(run_input={
# STATCODE or STATCODE/ITEMCODE1 — item codes differ per table
"statCodes": ["722Y001/0101000", "901Y009/0"], # base rate, CPI
"frequency": "M",
"startPeriod": "202401",
"endPeriod": "202610",
})
rates = pd.DataFrame(client.dataset(run["defaultDatasetId"]).iterate_items())
print(rates.pivot(index="period", columns="stat_code", values="value").tail(3))
For daily KRW/USD use "statCodes": ["731Y001/0000001"], "frequency": "D". All tables in one run share the
frequency. Not sure about codes? Run the actor with "mode": "items" and a table code to list its items.
3. From an AI assistant (MCP)
Apify's MCP server (https://mcp.apify.com) exposes public actors as tools for Claude Desktop, Claude Code, Cursor,
VS Code and other MCP clients. After adding it, a prompt like this works:
Use the Apify actor
kdata-hub/kosis-korea-statsto get South Korea's monthly CPI and unemployment rate
(tables101/DT_1J22003and101/DT_1DA7001S) from 2024-01 to 2026-10 and chart them.
Because rows are already in English with units and a source URL, the assistant can answer and cite in one step.
Ready-made examples (one click)
- Korea CPI, monthly — https://apify.com/kdata-hub/kosis-korea-stats/examples/korea-cpi-monthly
- Korea unemployment rate, monthly — https://apify.com/kdata-hub/kosis-korea-stats/examples/korea-unemployment-rate-monthly
- Bank of Korea base rate, monthly — https://apify.com/kdata-hub/ecos-korea-stats/examples/korea-base-rate-monthly
- KRW/USD, daily — https://apify.com/kdata-hub/ecos-korea-stats/examples/krw-usd-exchange-rate-daily
Cost and limits
- Pay per result: $2 per 1,000 rows. A 34-month CPI series filtered to the headline index is 34 rows (about $0.07); a whole multi-item table costs more, so filter items when you can.
- Up to 20 tables per run;
maxRecordscaps rows per table and the run reports truncation. - The numbers are the official KOSIS / ECOS values, not estimates; if the official source has a gap, the row says
so (
value_status) instead of inventing a value.
When not to use this
If you only need one or two headline series and are happy with a monthly lag, a free aggregator may be enough.
This is for when you want any Korean table, in English, with sources, from code or an AI tool — without
setting up Korean accounts and translation yourself.
Data: Statistics Korea (KOSIS) and Bank of Korea (ECOS). The actors are independent tools that call the official
open APIs; they are not affiliated with or endorsed by those institutions.
Top comments (0)