Analyze Korean Stocks with 6 Master Strategies

korea-stock-mcp is a conservative, data-honest MCP server for analyzing Korean-listed (KOSPI/KOSDAQ) equities from DART electronic-disclosure financial statements and KRX market data only. If a figure isn't available it reports null with a reason instead of substituting a default, and it deliberately returns a valuation range under stated assumptions rather than a single target price or a buy/sell call. Five tools are exposed: search_company (find a listed company by name or ticker), get_quote (price, market cap, 52-week high/low, volume, plus a market-cap consistency check against price times share count), get_financials (actual annual revenue, operating profit, and cash flow for 2-10 years, preferring consolidated over separate statements), get_valuation (PER/PBR multiples plus a pessimistic/neutral/optimistic value range with growth-rate and multiple assumptions stated), and get_risk_flags (six automated financial red-flag checks plus disclosure-risk keyword highlighting, over a configurable disclosure-lookback window). Every response carries an as_of date and a data_source field.

Use it when you need Korean-equity financials, a quote with cap-consistency checking, a scenario-based valuation range, or a financial/disclosure risk scan - through Claude, Cursor, or any MCP client. Not for non-Korean equities, real-time intraday data (KRX pricing is a daily snapshot, not live), quarterly financials (DART only provides annual reports through this server), or investment advice - the server's own disclaimer states its output is an arithmetic derivation of disclosure and price data, not investment advice, and that buy/sell/position decisions are entirely the user's responsibility.

Built on FastMCP 3 (Python 3.12+), running as a stateless remote server. Financial data comes from the DART OpenAPI (annual reports only, no quarterly data); price and market-cap data comes from KRX via FinanceDataReader (daily snapshots); free cash flow is an approximation (operating cash flow minus capex) that excludes intangible-asset and lease investment. Operational safeguards include global rate limiting, DART-quota-protecting caching, and serialized KRX requests to avoid IP blocks; the hosted deployment also auto-detects and throttles abusive request patterns (per-IP limits with escalating temporary blocks).

The hosted server needs no API key (it only serves public data): add https://korea-stock-analyzer-mcp-production.up.railway.app/mcp as a custom connector in claude.ai, or via Claude Code with claude mcp add --transport http korea-stock https://korea-stock-analyzer-mcp-production.up.railway.app/mcp, or as an "http"-type entry in Claude Desktop/Cursor's mcp.json. Self-hosting (e.g. on Railway, which auto-detects the Dockerfile) requires a free DART_API_KEY from opendart.fss.or.kr and a FASTMCP_HTTP_ALLOWED_HOSTS allowlist; running locally uses uv sync then DART_API_KEY=<key> uv run uvicorn app.main:app --app-dir src --port 8000, with uv run pytest for the unit/boundary suite and uv run pytest -m live for tests against the real KRX/DART APIs. The project is Apache-2.0 licensed.

Analysts and researchers who want Korean KOSPI/KOSDAQ financials and valuation ranges grounded strictly in actual DART/KRX data - with every gap reported as a gap rather than papered over - through Claude or another MCP client, for research rather than as a source of investment recommendations.