ko-edgar

Official Python SDK for ko.io — source-traced SEC & market data for AI agents and quants. 13F institutional holdings (85M+ rows, 2013→today), insider trades, Congress trading, crypto ETF exposure, macro indicators, and a white-labeled EDGAR filings gateway.

pip install ko-edgar

Quickstart

Works out of the box in keyless demo mode — no signup required:

from ko_edgar import KoClient

ko = KoClient()  # demo mode; or KoClient(api_key="ko_live_...")

# Who is Berkshire Hathaway holding right now?
for holding in ko.institutions.holdings("1067983", per_page=10):
    print(holding["ticker"], holding["holding_value"])

# Which institutions hold NVDA?
for holder in ko.stocks.holders("NVDA"):
    print(holder["name"], holder["shares_held"])

# What did Congress trade recently?
for trade in ko.congress.trades(sort="recent", per_page=10):
    print(trade["member_name"], trade["ticker"], trade["transaction_type"])

Get a free API key (200 calls/day, no credit card) at ko.io/console, then:

ko = KoClient(api_key="ko_live_...")   # or set KO_API_KEY env var

Async

from ko_edgar import AsyncKoClient

async with AsyncKoClient() as ko:
    result = await ko.stocks.activity("NVDA", quarters=8)

Pagination

from ko_edgar import KoClient, paginate

ko = KoClient()
for trade in paginate(ko.congress.trades, ticker="NVDA", per_page=100):
    ...

Results and errors

Every method returns an ApiResult: iterate it for rows, or use result.data / result.meta for the raw envelope. result.truncated tells you when the plan's row cap trimmed the response.

from ko_edgar import KoClient, PlanRequiredError, RateLimitError

ko = KoClient()
try:
    yields = ko.macro.treasury_yields(days=90)   # requires Pro
except PlanRequiredError:
    print("macro data needs a Pro plan → https://ko.io/pricing")
except RateLimitError as e:
    print(f"quota resets in {e.retry_after}s")

Escape hatch

Any /api/v1 endpoint not covered by a typed method:

ko.get("/api/v1/exec-compensation", ticker="AAPL", ceo_only=True)

Links

• Docs: https://ko.io/docs
• MCP server (same data inside Claude/Cursor/any agent): https://ko.io/mcp
• Repository: https://github.com/SharpLu/ko-mcp

MIT licensed.