FinBridge MCP Server

Explore one DART filing by receipt number while preserving section and table structure — the main body by default, or one of its attachments (audit reports carry the consolidated statements; the main body carries the business description). Use get_dart_filings first to obtain rcept_no. Start with action='overview', then select a section or table instead of requesting a long flattened filing. The overview lists the archive's entries (main body + attachments with their first heading); pass entry='<name>' to explore an attachment. Args: - rcept_no: 14-digit DART receipt number - entry: archive entry name from the overview's entries list (default: the main body) - action: overview (default) | section | table | compare_tables | compare_sections - compare_rcept_no and compare_index: second filing and its selected table/section index for comparison; table_index/section_index selects the first. Use both overviews first. Only selected content is compared, not entire filings. - section_index or section_query: one section selector for action='section' - table_index: 1-based table selector for action='table' - max_chars: section text cap, 1,000-100,000 (default 30,000) - max_rows/max_columns: selected table caps (defaults 50/30) - overview_offset/overview_limit: paginate both section and table summaries with the same 0-based slice (defaults 0/20, max 100) - response_format: markdown or json Returns source evidence for the OpenDART document entry and character range. Tables retain row/column order, cell text, rowspan/colspan and nearby reported unit labels. Zero and negative strings are not discarded. Separate HWP/PDF files are not fetched or rehosted. This tool does not claim an XBRL presentation/calculation hierarchy. Use get_dart_financials for normalized figures, reported account groups and known statement sum checks.

List corporate disclosure filings from DART, optionally filtered by company, date range, and disclosure type. Report names are in Korean. Not this tool for: US filings (get_edgar_filings), a cross-market feed already stored here (get_disclosure_feed), or major-event reports specifically (get_dart_major_events, a narrower slice of this one). Args: - corp: optional — Company: Korean name (e.g. '삼성전자'), 6-digit stock code (e.g. '005930'), or 8-digit DART corp_code (e.g. '00126380'). Omit for a market-wide list. - from / to: YYYY-MM-DD (default: last 90 days) - type: DART pblntf_ty — A=periodic reports(정기공시), B=major events(주요사항보고), C=securities issuance(발행공시), D=ownership/stake(지분공시), E=other(기타공시), F=external audit(외부감사관련), G=funds(펀드공시), H=asset securitization(자산유동화), I=KRX disclosures(거래소공시), J=fair trade(공정위공시) - limit: results per page, 1-100 (default 20); page: page number (default 1) Returns: {total, page, filings: [{rcept_no, corp_name, report_nm, flr_nm, rcept_dt, url}]} — url opens the filing in the DART viewer. Pass rcept_no to get_dart_document to explore its primary body by outline, section, or table. Examples: - {corp: '삼성전자'} -> Samsung filings in the last 90 days - {type: 'A', from: '2026-03-01', to: '2026-03-31'} -> March periodic reports market-wide Use to track what a company disclosed. For major events with keyword filtering use get_dart_major_events. Errors: no filings in range (DART status 013) -> widen dates or drop filters; unknown company -> search_dart_company.

Fetch financial statements of a Korean company from OpenDART (fnlttSinglAcntAll: full single-company statements) and normalize them to standard metrics. Amounts are raw KRW (no scaling); EPS is KRW per share. Not this tool for: US statements (get_edgar_financials), a KR-vs-US pair on one screen (compare_financials_kr_us), or ranking many companies at once (query_db, which reads the stored table and covers KR/US/TW/JP/EU). This tool requests one Korean company's statements through the DART adapter; results may be reused from a process-local cache for up to 24 hours. When FinBridge's own database already has this exact company/year/report/basis with account-level detail and no newer filing receipt, it answers from the database instead of calling DART (served_from: finbridge_db) — saved to conserve the shared daily DART quota; a 'statement'/'account_query' filter always goes live since the DB keeps only the top reported rows. data_as_of.generated_at is response creation time, not source retrieval time. Peer comparisons use separately dated database snapshots. Args: - corp: Company: Korean name (e.g. '삼성전자'), 6-digit stock code (e.g. '005930'), or 8-digit DART corp_code (e.g. '00126380') - year: business year 2015-2026 (default: last year). Annual reports are filed ~March of the following year (FY2025 filed 2026-03). - report: 'annual' | 'q1' | 'half' | 'q3' (default 'annual') - fs: 'consolidated' | 'separate' (default 'consolidated'). If consolidated statements do not exist, automatically retries separate and says so in notes. - statement: optional BS/IS/CIS/CF/SCE account-group filter - account_query: optional case-insensitive account name or account_id substring - account_limit: returned account rows, 1-200 (default 40) - as_of: optional YYYY-MM-DD. Adds point_in_time: the version of this period/basis that was public on that day. - response_format: 'markdown' (default, tables) or 'json' (compact) Returns structured {normalized, accounts}: - normalized: {company:{name,id,ticker}, basis, periods:[{period, fiscal_year, currency:'KRW', metrics:{revenue, gross_profit, operating_income, net_income, eps_diluted, assets, liabilities, equity, cash_and_equivalents, operating_cash_flow}}], notes}. Annual reports include the prior-year comparative as a second period. - accounts: selected reported statement rows with statement, account id/detail, reported order, currency, current and previous amounts. - statement_groups: counts and reported order ranges by BS/IS/CIS/CF/SCE. These are flat DART groups, not an inferred XBRL hierarchy. - sum_checks: known balance-sheet equations with pass/fail/not_tested, operands, currency, tolerance and receipt-number evidence. - preserved_revisions: bounded numeric filing versions for the exact database company, requested period and actual statement basis. Includes provider, metric units, publication-day evidence, latest observation, unknown first-seen state, and explicit comparison uncertainty. - revision_links: original filing <-> amendment filings for this period/basis, linked only by DART report names ([기재정정]/[첨부정정] prefix, same title, one original, later receipt numbers). restated_items lists each metric whose value changed with original_value, amended_value, both receipt numbers, both publication dates and both DART URLs. Exact 1,000^k ratios are flagged as suspected unit errors, not restatements. Unlinkable cases return a reason instead of a guess. - point_in_time (only with as_of): the last preserved version with publication_date <= as_of (no look-ahead; latest-value fallback is never used). status found | uncertain (an amendment public by as_of has no preserved values) | not_yet_filed | original_not_preserved | no_versions | unavailable. - earnings_disclosures (only when stored filings exist for this period): figures the company itself disclosed, next to its periodic-report figures. preliminary_vs_reported = the last 영업(잠정)실적 filing published on or before the periodic report vs that report's original values (difference, difference_pct, within_rounding = within half the preliminary table's unit; window_ambiguous/scale_suspect assert nothing), with every preliminary version, its amendment link (only when exactly one earlier original exists) and unreadable filings. outlook_vs_reported = single-number items from 영업실적등에대한전망/장래사업ㆍ경영계획 for this exact period with achievement_pct = reported ÷ disclosed × 100, or a status explaining why nothing is computed (basis_not_stated, definition_may_differ, period_not_reported_directly …). history = counts of those filings and their amendments since recorded_from. Facts only — no forecast, estimate or recommendation. Examples: - {corp: '삼성전자', year: 2024} -> FY2024+FY2023, revenue ~3.0e14 KRW - {corp: '005930', year: 2025, report: 'q1'} -> Q1 2025 (3-month income-statement figures) - {corp: '010130', year: 2024, as_of: '2025-06-30'} -> FY2024 as known on 2025-06-30 (before the 2026-08-13 amendment) Use for KR-listed company fundamentals. Don't use for US companies (get_edgar_financials) or for filings text (get_dart_filings). Errors: 'no data' (DART status 013) -> try another year/report or fs='separate'; unknown company -> run search_dart_company first.

Korean insider transactions for a listed KR company, from DART's 임원ㆍ주요주주 특정증권등 소유상황보고서 (elestock) — the Korean equivalent of SEC Form 4. Includes a buy-vs-sell summary and an optional buy/sell filter. Not this tool for: US insiders (get_edgar_insider_trades) or institutional managers, which are a different kind of holder entirely (get_edgar_13f). Buy vs sell is the SIGN of the reported share change (증감수): positive = 취득 (acquire / buy), negative = 처분 (dispose / sell). Insider BUYING is a stronger sentiment signal. Args: - company (required): KR 6-digit stock code (e.g. '005930'), company name, or 8-digit DART corp_code - limit: number of most-recent reports to return, 1-100 (default 20) - tx_type: 'all' (default) | 'buy' (share change > 0) | 'sell' (share change < 0) - response_format: 'markdown' (default) or 'json' Returns: {company:{corp_code, corp_name}, tx_type, summary:{buys:{count,shares}, sells:{count,shares}}, count, trades:[{filedAt, reporter, position, registered_exec, major_shareholder, change, shares_after, change_rate}], notes}. summary totals cover the whole fetched set regardless of the filter. Important: this elestock summary API does not provide transaction dates or prices. filedAt is the filing date, not the transaction date. The underlying filing may contain these details; use get_dart_document with the returned rcept_no to inspect its tables. Examples: - "삼성전자 임원 매수" -> {company:'005930', tx_type:'buy'} - "SK하이닉스 내부자 매도 최근" -> {company:'000660', tx_type:'sell'} Use when: monitoring KR officer / major-shareholder buy/sell activity. For US insiders use get_edgar_insider_trades. For institutional holdings use get_edgar_13f. Errors: unknown company -> use search_dart_company; a filter with no matches returns count 0 (not an error).

List major-event disclosures (주요사항보고서, DART type B): capital increases, mergers, convertible bonds, treasury stock, bankruptcy, lawsuits, etc. Optionally filter report names with a regex. Not this tool for: the full disclosure list or other report categories (get_dart_filings — periodic reports, securities issuance, ownership, KRX notices), US 8-K events (get_edgar_filings), or the cross-market stored feed (get_disclosure_feed). What this adds over get_dart_filings type='B': a 'kinds' regex over Korean report names (e.g. '증자|합병|전환사채'), a 180-day default window tuned for event scans, and matched-count totals — so use it when the question is 'which companies announced X', not 'what did company Y file'. Args: - corp: optional — Company: Korean name (e.g. '삼성전자'), 6-digit stock code (e.g. '005930'), or 8-digit DART corp_code (e.g. '00126380'). Omit for market-wide events. - from / to: YYYY-MM-DD (default: last 180 days) - kinds: optional JavaScript regex matched against the Korean report name, e.g. '증자|합병|전환사채' (capital increase | merger | CB) or '자기주식' (treasury stock). Filtering runs over every event in range when FinBridge's index covers the window, otherwise over the most recent 100 events DART returns. - limit: max results, 1-100 (default 20) Returns: {total, page, filings: [{rcept_no, corp_name, report_nm, flr_nm, rcept_dt, url}]} — same shape as get_dart_filings. When kinds is given, total = matched count within the scanned window. Examples: - {corp: '삼성전자', kinds: '자기주식'} -> Samsung treasury-stock decisions in the last 180 days - {kinds: '유상증자', from: '2026-01-01', to: '2026-06-30'} -> market-wide rights offerings in H1 2026 Use when: event-driven screening (rights offerings, mergers, CBs, treasury stock, lawsuits) market-wide or for one company. For every filing category, or when you already know the report you want, use get_dart_filings. Errors: no events in range (DART status 013) -> widen dates; invalid kinds regex; unknown company -> search_dart_company.

Search companies registered with DART, South Korea's corporate disclosure system, by name, 6-digit stock code, or 8-digit DART corp_code. Returns the corp_code required by the other dart_* tools. Not this tool for: US registrants (use search_edgar_company). Japan, Taiwan and Europe have no search tool — reach them through query_db on the companies table. Args: - query: company name in Korean ('삼성전자') or English ('Samsung Electronics'), 6-digit KRX stock code ('005930'), or 8-digit corp_code - listed_only: restrict to KRX-listed companies (default true). Set false to include ~90k unlisted entities. - limit: max results, 1-50 (default 10) Returns: {count, companies: [{corp_code, corp_name, corp_name_en, stock_code}]} — corp_name is the Korean name DART registers, corp_name_en the registered English name where the company files one (about three quarters of Korean issuers). stock_code is null for unlisted companies. Match priority: exact stock code > exact Korean name > listed Korean partial > unlisted Korean partial > English name. English matching reads our own company table, so it covers issuers whose English name we hold, not every DART entity. Examples: - {query: '삼성전자'} -> corp_code 00126380, stock_code 005930 - {query: '카카오', listed_only: false} -> listed 카카오 plus unlisted same-name entities - {query: 'Samsung Electronics'} -> 삼성전자 (matched on the registered English name) Use when you need a corp_code or must disambiguate similar names. Don't use for US companies (use search_edgar_company). Errors: DART_API_KEY not configured; no match returns count 0 (not an error).

Quarter-end institutional manager holdings reconstructed from SEC 13F-HR and 13F-HR/A (RESTATEMENT or NEW HOLDINGS), for the latest reportDate in recent submissions. Pass a manager name or CIK, not an issuer ticker. top=1–50 (default 20) limits displayed rows. response_format=markdown or json. Returns reported security rows with separate class, PUT/CALL, SH/PRN, discretion and other-manager fields; no ticker mapping or reverse ownership lookup. Values normalize each filing to USD using the 2023-01-03 filing-date boundary. Missing numbers stay null. Duplicate/shared reporting, incomplete amendment chains and confidential omissions withhold totals/weights. sources carries accession, primary/table URLs, report/filed/acceptance/fetch times and amendment evidence. prior_period is a distinct reportDate; no prior omits changes. Unverified corporate actions, missing rows, options/principal and incomplete reports withhold change signals with explicit comparison_exclusions. Empty changes does not establish no activity. 13F covers disclosed Section 13(f) positions, not a complete/current portfolio; 45 days is a filing deadline, not a freshness guarantee. Not investment advice.

List a US company's recent SEC filings (10-K, 10-Q, 8-K, S-1, proxy statements, Form 4, ...) from the EDGAR submissions index. Returns metadata and document URLs only — it does NOT download filing contents; fetch the returned url yourself for the document text. Not this tool for: Korean filings (get_dart_filings) or a cross-market feed already stored here (get_disclosure_feed). Args: - company (required): ticker / company name / CIK - forms: optional form-type filter, e.g. ['10-K'] or ['10-K','10-Q','8-K'] (exact match, case-insensitive) - from / to: optional YYYY-MM-DD filing-date range - limit: max rows, 1-50 (default 20) Returns: {company:{cik, name, ticker}, count, filings:[{form, filingDate, accessionNumber, primaryDocument, items?, url}], notes?}. 8-K rows include 'items' (e.g. '2.02,9.01' = results of operations + exhibits). Coverage = the latest ~1000 filings per company. Examples: - "Apple's latest annual report" -> {company:'AAPL', forms:['10-K'], limit:1} then fetch the url - "Tesla 8-Ks this year" -> {company:'TSLA', forms:['8-K'], from:'2026-01-01'} Use when: you need filing dates, document links, or 8-K event items for a US company. Don't use for: Korean disclosures (get_dart_filings) or filing full-text search across all companies. Errors: unknown company -> use search_edgar_company; an empty result usually means the form/date filter is too narrow for the ~1000-filing window.

Normalized annual (10-K) or quarterly (10-Q) financial statements for a US company, from SEC EDGAR XBRL company facts (US-GAAP). Values are raw USD (not scaled); eps_diluted is USD per share. Not this tool for: Korean statements (get_dart_financials), a KR-vs-US pair on one screen (compare_financials_kr_us), or ranking many companies at once (query_db, which reads the stored table and covers KR/US/TW/JP/EU). This tool requests one US company's SEC XBRL company facts through the source adapter; results may be reused from a process-local cache for up to 24 hours. data_as_of.generated_at is response creation time, not source retrieval time. Peer comparisons use separately dated database snapshots. Args: - company (required): ticker / company name / CIK (e.g. 'AAPL', 'Microsoft', '789019') - freq: 'annual' (default, from 10-K) or 'quarterly' (discrete Q1-Q3 from 10-Qs; Q4 is not reported separately) - periods: how many most-recent periods, 1-12 (default 3) - metrics: optional subset of [revenue, gross_profit, operating_income, net_income, eps_diluted, assets, liabilities, equity, cash_and_equivalents, operating_cash_flow] (default all) - response_format: 'markdown' (default) or 'json' Returns NormalizedFinancials: {company:{name, id(CIK), ticker}, basis:'US-GAAP (10-K)', periods:[{period:'FY2024', fiscal_year, end, currency:'USD', metrics:{revenue, net_income, ...}}], notes}. periods are most-recent first; fiscal_year = calendar year of the period end date. Examples: - "Apple's revenue and net income for the last 3 years" -> {company:'AAPL', metrics:['revenue','net_income']} - "MSFT last 4 quarters" -> {company:'MSFT', freq:'quarterly', periods:4} Use when: you need US-GAAP fundamentals for a US-listed company. Don't use for: Korean companies (get_dart_financials), stock prices, or IFRS 20-F foreign private issuers (not supported). Errors: unknown company -> use search_edgar_company first; companies without us-gaap XBRL facts (funds, 20-F filers) return an error explaining why.

Latest insider transactions for a US company, parsed from SEC Form 4 filings, with a buy-vs-sell summary and an optional buy/sell filter. Each trade lists the reporting insider, their relationship, and non-derivative (common stock) transactions. Not this tool for: Korean insiders (get_dart_insider_trades) or institutional managers, which are a different kind of holder entirely (get_edgar_13f). Insider BUYS (open-market purchases, code P) are a stronger sentiment signal than sells (code S), which happen for many reasons (diversification, taxes). Use tx_type to monitor one side. Args: - company (required): ticker / company name / CIK - limit: number of most-recent Form 4 filings to parse, 1-25 (default 10) - tx_type: 'all' (default) | 'buy' (code P purchases only) | 'sell' (code S sales only) Returns: {company:{cik, name, ticker}, tx_type, summary:{buys:{count,shares,value}, sells:{count,shares,value}}, count, trades:[{filedAt, owner, relationship, url, transactions:[{date, code, shares, price_per_share, acquired_or_disposed, shares_owned_after}]}], notes}. summary totals cover the whole fetched window regardless of the filter; value = shares x price where a price is reported. Transaction codes: P=open-market purchase, S=open-market sale, M=option exercise, F=shares withheld for tax, A=award/grant, G=gift. acquired_or_disposed: A=acquired, D=disposed. Examples: - "insider BUYING at Apple" -> {company:'AAPL', tx_type:'buy'} - "recent insider SELLING at Nvidia" -> {company:'NVDA', tx_type:'sell'} - "all TSLA insider activity, more history" -> {company:'TSLA', limit:25} Use when: monitoring insider buy/sell activity (officers, directors, 10% owners) for a US-listed company. Larger 'limit' widens the time window. Don't use for: institutional holdings (use get_edgar_13f), Korean companies, or derivative-only detail (option grids are skipped). Errors: unknown company -> use search_edgar_company; a filter with no matching transactions returns count 0 (not an error); unparseable Form 4 XMLs are skipped and counted in notes.

Search SEC EDGAR registrants (US-listed companies) by ticker, company name, or CIK. Returns the 10-digit zero-padded CIK needed by the other edgar_* tools. Not this tool for: Korean companies (use search_dart_company). Japan, Taiwan and Europe have no search tool — reach them through query_db on the companies table. Args: - query (required): ticker ('AAPL', 'BRK-B' or 'BRK.B'), company-name fragment ('Berkshire'), or CIK number ('320193') - limit: max results, 1-50 (default 10) Returns: {count, companies: [{cik, ticker, title}]} ranked exact-ticker > exact-name > prefix > substring. Examples: - "find Apple's CIK" -> {query: 'AAPL'} - "companies named Berkshire" -> {query: 'Berkshire', limit: 5} Use when: you need a CIK or to disambiguate a company name before calling get_edgar_financials/filings/insider_trades (those also accept tickers directly, so for an exact ticker you can skip this step). Don't use for: Korean companies (use search_dart_company) or private companies not registered with the SEC. Errors: no match -> error suggesting a shorter name fragment; only SEC registrants with a listed ticker are searchable.

Return the holdings this user has already registered in FinBridge with import_portfolio — listed stocks plus cash, crypto, ETF and physical assets — valued at the latest prices, with return and asset allocation. A user who has registered nothing gets an empty list. Not this tool for: analysing or valuing a company (get_valuation), prices (get_stock_prices), or finding companies (search_dart_company / search_edgar_company). It reads only what this user stored, so it knows nothing about a company they do not hold. Crypto and ETFs use live ccxt quotes, stocks the latest close in the database, cash and physical assets the registered amount. allocation is aggregated per currency; combined converts everything to KRW using an ECB-derived USD/KRW rate.

List the append-only portfolio snapshots recorded for this user. Each import_portfolio call that actually changes the stored holdings/assets (and each restore_portfolio_snapshot) appends one snapshot of the whole portfolio at that moment — quantities, average prices, symbols and asset classes, never market prices or computed valuations (those are recomputed fresh whenever needed). Identical resubmissions do not create a duplicate entry. Use this to see when the portfolio changed, then restore_portfolio_snapshot to undo a bad import.

Compare annual financial statements of a Korean listed company (source: OpenDART, K-IFRS) and a US listed company (source: SEC EDGAR, US-GAAP) side by side, with KRW values converted to USD using Federal Reserve H.10 annual-average exchange rates. Args: - kr_company: Korean company name / 6-digit stock code / DART corp_code (e.g. '삼성전자', '005930') - us_company: US ticker / name / CIK (e.g. 'AAPL', 'Apple') - years: number of recent fiscal years, 1-5 (default 3; free plan up to 4) - metrics: subset of [revenue, gross_profit, operating_income, net_income, eps_diluted, assets, liabilities, equity, cash_and_equivalents, operating_cash_flow] - response_format: 'markdown' (default) or 'json' Returns per-metric, per-year rows: {fiscal_year, kr_krw, kr_usd, us_usd, ratio_kr_over_us} plus the FX rates used, accounting-basis caveats, sources (the DART receipt or SEC accession behind each year's figures, with links) and data_as_of. Examples: - "삼성전자 vs Apple 최근 3년 매출·영업이익 비교" -> {kr_company:'삼성전자', us_company:'AAPL', metrics:['revenue','operating_income']} - Don't use for quarterly data (annual only) or non-KR/US companies. Errors: unknown company names suggest using search_dart_company / search_edgar_company first.

Recent regulatory disclosures from the local finbridge database (filings table, refreshed nightly + intraday for KR), newest first — positioned as a faster-than-news primary source. By default returns only MATERIAL filings: US Form 8-K (current reports) and KR 주요사항보고서 (major events: capital raises, M&A, convertible bonds, buybacks, etc.). Args: - market: 'kr' (DART), 'us' (EDGAR), or 'all' (default) - company: optional — restrict to one company (US ticker, KR 6-digit code, or name). Takes priority over watchlist. - material_only: default true (8-K / KR type-B only); false = all filing types - forms: optional explicit form_type filter (e.g. ['10-K','8-K'] or ['A','B']); overrides material_only - days: look-back window in days, 1-120 (default 14); or use from/to - from/to: optional explicit YYYY-MM-DD range (overrides days) - limit: 1-100 (default 30); response_format: 'markdown'|'json' - brief: default true — attach the plain-language fact summary (headline_en/headline_ko/facts_en/facts_ko/company_en/event_type) where one exists. Rows without a summary are unaffected. The markdown rendering shows the English side; both sides stay in the JSON. - watchlist: default false — restrict to the authenticated account's watchlist companies instead of all companies. Needs a personal API key (not the admin token). Ignored when 'company' is also given. Returns: {count, market, since, rows:[{source, company_name, form_type, title, filed_date, receipt_no, url, items?, summary_status?, headline_en?, headline_ko?, facts_en?, facts_ko?, company_en?, event_type?}]}. 'items' (8-K item codes) is included when available. Summary fields are present only for filings that have one. - summary_status (with brief:true): 'ok' = summary attached; 'withheld_review' = a summary exists but failed our accuracy/wording checks and is withheld pending review (title+URL only — read the filing); 'not_generated' = no summary for this filing yet. Note: with brief:true, a filing that only exists as a summary (not yet in our filings index) can still appear — such rows omit form_type/items and, for non-DART sources, url. Examples: - Latest US material events this week: {market:'us', days:7} - Samsung's recent major-event filings: {company:'005930', material_only:true, days:90} - All of a company's filings: {company:'AAPL', material_only:false} - My watchlist's recent disclosures with summaries: {watchlist:true} Use when: scanning for catalysts / breaking corporate events, or one company's recent filings. Don't use for filing BODIES (open the url) or for financial statement values (get_dart_financials / get_edgar_financials / query_db). Notes: Filing metadata only; bodies are at the linked source URLs. Not investment advice. Errors: empty result is not an error (count 0).

Store the structured holdings explicitly entered by the user in their FinBridge portfolio. Uploaded files, screenshots, chat history and extracted file content are not supported sources for this connector. Accepts listed stocks (KR/US/TW/JP) as well as cash, crypto (BTC etc.) and physical assets (gold): stocks are matched against the database, crypto and gold (PAXG) get live ccxt quotes, cash and physical assets are stored at the given value. For ETFs or foreign products not in the database, pass value directly. If the user specifies an asset class, pass asset_class as well (cash|bond|physical|growth|dividend|crypto|other; Korean labels 현금|채권|현물|성장주|배당주|가상자산|기타 are accepted). Registered listed stocks are also added to the watchlist automatically. Use when: the user explicitly enters what they hold and wants it stored for get_portfolio. There is no per-holding edit or delete tool: to change or remove holdings, re-import the complete corrected list with replace=true (replace=false only adds/updates the rows given). Not this tool for: the watchlist (manage_watchlist — companies followed, no quantities), valuing a company (get_valuation), or reading what is already stored (get_portfolio). IMPORTANT — read the response before telling the user you are done: 1. Confirmation gate: if the user already has a stored portfolio, this call returns `preview:true` with a `changes` diff (added/removed/changed) and does NOT save anything, unless you pass confirm=true. Show the diff to the user — call out `changes.removed` especially: if the submission was only part of their holdings, those positions will look fully sold. Only pass confirm=true after the user has seen and accepted the diff (skip this if `get_portfolio` was empty to begin with — there is nothing to compare against). 2. Missing fields: each saved row reports `missing_fields` (commonly `acquired_on`, since brokerage statements rarely include it) and unresolved symbols appear in `needs_input` — ambiguous names/codes list `candidates` (do not guess one), unmatched symbols need a ticker/code or a `value`. Rows with missing fields ARE saved (never block on incompleteness) — ask the user for the missing pieces and call this tool again for just those rows to fill them in. 3. Cash: check the `cash` field. If `status` is `not_asked`, the stored portfolio has no cash balance on file (this is different from a confirmed zero) — ask the user for their cash balance (amount or % of the portfolio) and its currency, then import it as one more holding: `{symbol:'현금', value, currency, asset_class:'현금'}`. Without it, equity weights in get_portfolio are overstated. 4. Always show the user the per-row summary (name/qty/avg_price recognized) before treating the import as authoritative — a misread quantity produces a wrong return. If something was saved wrong, undo it with restore_portfolio_snapshot (defaults to the snapshot right before this one).

Taiwan insider share-transfer filings from TWSE (上市) and TPEx (上櫃), served from the local finbridge database. ⚠These are PRE-ANNOUNCEMENTS, not executed trades. Taiwan requires directors, supervisors, managers and 10% shareholders to declare a transfer BEFORE selling (內部人持股轉讓事前申報). There is no "sold" table at all — a declaration says what someone intends to transfer and by when. What does exist is the opposite: an 未轉讓 (not-transferred) table listing declarations whose window expired without a sale, with the filer's stated reason. This tool returns both. This is why it is a separate tool from get_dart_insider_trades (Korea) and get_edgar_insider_trades (US Form 4), which report trades that already happened. Do not compare the numbers across those tools as if they were the same event. ⚠Coverage is short and has permanent holes. The upstream endpoints publish only the CURRENT day's table — there is no historical query — so our history starts when we began collecting and any day the collector missed is unrecoverable. The response's coverage.first_report_date and coverage.days say exactly how much history exists; "no rows" for an earlier date means we never had it, not that nobody filed. ⚠No rankings or aggregates in this version (no "most-sold-by-insiders this month"). A few days of snapshots is not a sample. Units are SHARES (股) — not the thousands of shares used by Taiwan margin data. No monetary conversion is done: planned shares times a closing price is not a transaction value. Role, method and reason strings are returned in the original Chinese so they can be checked against the source. Args: - company: optional filter — TW 4-digit code, '2330.TW', 'tw:2330', or the company name - from / to: report_date range (YYYY-MM-DD) - role: substring of the declarant's role in Chinese (董事 / 監察人 / 經理人 / 大股東 / 法人董事代表人) - min_shares: minimum planned_shares (applies to the transfer table only) - include_untransferred: also return expired declarations that were not acted on (default true) - limit: 1-200 (default 50); response_format: 'markdown' | 'json' Returns: {market, company?, range, transfers: [{report_date, company, declarant_role, declarant_name, method, period_from, period_to, planned_shares, daily_max_shares, transferee, held, planned, after}], untransferred: [{report_date, company, declarant_role, declarant_name, untransferred, held, declared, reason}], coverage: {first_report_date, last_report_date, days}, notes, data_as_of, page_url} Use when: someone asks what Taiwanese insiders have declared they intend to sell, or whether a declared sale actually went through (an 未轉讓 row means it did not). Don't use it to claim an insider "sold" — this data cannot show that.

Add or remove a company from the user's watchlist. This changes saved preferences, not financial assets. Only remove deletes an entry. action='add' and action='remove' each take one symbol and are idempotent: adding a company already on the list leaves it there, removing one that is not on the list is a no-op. Both report the resulting list size. Not this tool for: holdings and cash (that is a portfolio — use get_portfolio / import_portfolio), or for any market data. The watchlist stores which companies the user follows, nothing about quantities, prices or returns.

Comparison references for one company across KR / US / TW / JP / EU. The default uses a sourced business theme or broad source classification and does not assert direct competition or add unrelated companies to fill the limit. Explicit rank='size' returns same-currency size references and does not assert an industry relationship. Also returns the company's business-segment revenue split where available (Japan from 有価証券報告書 XBRL, the US from SEC DERA financial-statement datasets; US segment names are usually end markets, not industries) — informational unless rank='segments'. Args: - company: US ticker ('AAPL'), KR 6-digit code ('005930'), TW/JP 4-digit code ('2330', '7203'), or a company name (local or English). - market: 'kr'|'us'|'tw'|'jp'|'eu' (optional) — disambiguates codes/names shared across markets (TW and JP both use 4-digit codes; 'eu' companies are addressed by ISIN). - limit: 1-10 peers (default 5). - same_market_only: true = restrict peers to the company's own market (default false — a KR chipmaker can sit next to a US one). - rank omitted = business-related references; 'size' = explicit size references with verified equal market-cap currency; 'segments' = rank by business-mix similarity — each company's segment revenue shares are mapped to standard industries (companies without segment data count as 100% their own industry) and compared by cosine similarity, ties broken by normalized size. Conglomerates (Sony: games/music/pictures/electronics/finance) then get conglomerate peers instead of whichever single bucket they were filed under. - response_format: 'markdown' (default) or 'json'. Returns: existing fields plus policy_version, purpose, insufficiency_reason; each peer also has selection_reason, comparison_role and evidence_status. Broad/theme rows are business-related references, not verified direct competitors. Explicit size rows are size-reference only. Examples: - {company:'7203'} -> Toyota + transportation-equipment peers, with its Automotive / Financial Services segment split - {company:'005930', same_market_only:true} -> Samsung Electronics + KR tech-hardware peers only - {company:'6758', rank:'segments'} -> Sony ranked against other multi-segment conglomerates by business mix Use when: building a comparison table or choosing competitors for a financial comparison. Don't use for strategy screens (screen_*) or for full financial statements (get_dart_financials / get_edgar_financials). Notes: curated business themes remain separate from official company-level classifications. Segment names may be geographic (Japan/Asia/USA) when a company defines its reportable segments by region. Unverified cross-currency absolute market-cap ordering is never used. Errors: unknown/ambiguous company -> candidate list; no classification -> basis_kind='size' with a note.

Run a single read-only SELECT query against the local finbridge database (ingested KR/US fundamentals, filings, daily prices). The statement must start with SELECT or WITH; multiple statements, PRAGMA, and any write/DDL keywords (INSERT/UPDATE/DELETE/DROP/ALTER/CREATE/ATTACH/...) are rejected. The query runs in a separate read-only process with SQLite authorization, a 2-second deadline, two concurrent queries per server process, and a 1 MB result budget. Free accounts cannot query raw history or history views; the latest-annual snapshot remains available. filing_briefs returns summary text only for published summaries; withheld rows (status='review') come back with the summary columns NULL. The escape hatch for questions no dedicated tool answers — Japan, Taiwan and Europe are largely reachable only this way. Prefer get_valuation or get_peers for ordinary single-company work (they handles per-market period and currency rules that a hand-written query will get wrong), and call get_db_schema first for the table shapes. Args: - sql: one SELECT (or WITH ... SELECT) statement. A single trailing ';' is tolerated. - limit: max rows returned, 1-500 (default 50) - response_format: 'markdown' (default, table) or 'json' (compact) Returns: {columns: [name], rows: [[cell, ...]], row_count, truncated} — truncated=true means more rows matched than 'limit'. Examples (v_financials / v_latest_annual views are the easiest entry points): - Largest companies by latest annual revenue: "SELECT name, ticker, stock_code, fiscal_year, revenue FROM v_latest_annual ORDER BY revenue DESC LIMIT 10" - Samsung Electronics annual trend: "SELECT fiscal_year, revenue, operating_income, net_income FROM v_financials WHERE stock_code = '005930' AND quarter = 0 ORDER BY fiscal_year DESC" - KR vs US company counts: "SELECT source, COUNT(*) AS n FROM companies GROUP BY source" - Recent Samsung Electronics closes: "SELECT date, close FROM prices_daily p JOIN companies c ON c.id = p.company_id WHERE c.stock_code = '005930' ORDER BY date DESC LIMIT 20" (prices_daily holds KR, US, TW; US history starts 2023-03-28) Use when: custom aggregation/joins over ingested data that no dedicated tool expresses. Don't use for anything that writes — it will be rejected. FinBridge has no real-time equity quote tool — equity prices here are end-of-day closes from the nightly ingest; the only live data is regulator filings (get_dart_filings / get_edgar_filings). Errors: non-SELECT input, ';' inside, or forbidden keywords -> rejected with the reason; unknown table/column -> SQL error with a hint to call get_db_schema first.

Roll the stored portfolio back to a past snapshot from get_portfolio_history, replacing ALL current holdings and assets with that snapshot's content. Use this to undo a bad import_portfolio call. Omit snapshot_id to restore the snapshot immediately before the current state (undo the last change). This action is itself recorded as a new snapshot — history is append-only, so restoring is itself reversible the same way. Tell the user what was restored (get_portfolio afterwards shows it valued at current prices).

Inspect the schema of the local finbridge database (SQLite with ingested KR/US company fundamentals, filings, and daily prices): tables, views, columns, per-table row counts (counted in the background and refreshed every 30 minutes; null with rows_note "counting…" right after a server start), and ready-to-run example queries for query_db. Read this before writing a query_db statement. It returns no company data itself — get_db_schema describes the tables, query_db runs the SELECT. Args: (none) Returns: {tables: [{name, columns: [{name, type}], rows}], views: [{name, columns: [{name, type}]}], examples: [sql_string]} Key objects: - companies: KR companies have source='dart' + stock_code (6-digit), US companies source='edgar' + ticker - financials: one row per company x fiscal_year x quarter (quarter=0 = annual); raw unscaled KRW/USD amounts - prices_daily: daily OHLCV per company_id - views v_financials (financials joined with company name/ticker/stock_code) and v_latest_annual (latest annual row per company) — prefer these in query_db - filing_briefs: filing summaries keyed by (source, ext_key). Summary text is returned only for published summaries: a row withheld by the review checks (status='review') keeps its keys and dates with headline/facts/company_en/event_type/guards/engine NULL — the summaries get_disclosure_feed reports as summary_status='withheld_review'. Examples: - Call before writing SQL for query_db, to learn table/column names. - Check row counts to see how much data the nightly ingest has loaded. Use when: preparing a query_db, or checking ingest coverage. Don't use for market data itself (get_stock_prices / get_valuation read the same tables with the right joins already done). FinBridge has no real-time equity quote tool — equity prices here are end-of-day closes from the nightly ingest; the only live data is regulator filings (get_dart_filings / get_edgar_filings). Errors: 'database has not been built yet' — the ingest pipeline has not run on the server.

Screen exchange-traded funds in the local finbridge database on the things that actually distinguish an ETF: premium/discount to NAV, fund size (AUM), the index it tracks, price momentum, and — for US funds — the audited calendar-year TOTAL return from the fund's own prospectus. Funds only. For an operating company use get_valuation (one company) or query_db (many at once). ⚠Funds file no annual financial statements, so the statement-based tools (get_dart_financials, get_valuation) have nothing to say about them — that is why they are screened here instead. Coverage differs by market and the response says so per row: - KR (1,170 listed ETFs): NAV, AUM (net assets, KRW), listed units and the tracked index come from the same daily feed as prices, 2020-01-02 onward. premium_pct is close/NAV-1 computed on the SAME day (mixing dates would be meaningless). - US (5,868 ETFs): no NAV or AUM source exists that we may redistribute, so those fields are null. Instead total_return_pct carries the fund's audited calendar-year total return (distributions reinvested) from SEC prospectus data — the only distribution-inclusive number available. ⚠ret_20d / ret_120d are PRICE returns in every market: ETF distributions are not in the daily bars, so income funds look worse than they were. For US funds compare against total_return_pct to see the gap. ⚠aum is in the listing currency (KRW today). Do not rank across markets on it. ⚠total_return_pct is pinned to ONE calendar year across all rows (reported as total_return_year), because prospectus refresh dates differ per fund — ranking a 2024 figure against a 2025 one would be a silently wrong table. Args: - market: 'kr', 'us', or 'all' (default) - min_aum: minimum net assets in listing currency (KR only; e.g. 100000000000 = 1,000억) - max_abs_premium_pct: keep funds trading within this |premium| of NAV, e.g. 0.5 - min_premium_pct: keep funds at or above this premium (negative values find discounts) - min_price, min_volume: liquidity floors (vol_avg20 is the 20-session average) - index_contains: substring of the tracked index name — 'TR' finds total-return index trackers, '코스피' finds KOSPI trackers - name_contains: substring of the fund name or ticker - total_return_year: calendar year for total_return_pct; omit for the best-covered year - sort_by: aum | premium | abs_premium | ret_20d | ret_120d | ret_250d | volume | total_return (default aum); order: 'asc'|'desc' (default desc) - limit: 1-100 (default 20); response_format: 'markdown'|'json' Returns: {count, market, total_return_year, sort_by, order, rows: [{market, symbol, name, as_of, close, nav, premium_pct, aum, index_name, ret_20d, ret_120d, vol_avg20, total_return_pct, total_return_period}]} Examples: - Large KR ETFs trading close to fair value: {market:'kr', min_aum: 100000000000, max_abs_premium_pct: 0.3, sort_by:'aum'} - KR ETFs at the deepest discount to NAV: {market:'kr', sort_by:'premium', order:'asc'} - KR trackers of a total-return index: {market:'kr', index_contains:'TR', sort_by:'aum'} - US ETFs by audited total return: {market:'us', sort_by:'total_return'} Use when: choosing or comparing funds. Don't use for stocks (get_valuation / query_db) or for a single fund's price history (get_stock_prices). Errors: 'database has not been built yet' — ingest has not run; an empty result is not an error (count 0).

Get daily OHLCV price history from the local finbridge database (populated by the nightly ingest jobs). Rows are returned newest first. Listed equities and ETFs. Japan and Europe carry no prices at all, and crypto is not served. Price coverage by market — we only store what we have redistribution rights to: - Korea (DART + Financial Services Commission): full daily history, corporate-action adjusted. SERVED. - Taiwan (TWSE OpenAPI, Open Government Data License): daily history. SERVED. - US (Databento EQUS.SUMMARY): daily history from 2023-03-28. SERVED. Split-adjusted; dividend-adjusted closes exist where SEC-reported dividends do (adj_close). - Japan: NOT served. EDINET publishes disclosure documents, not prices, so we hold Japanese filings and the company master but no quotes. Args: - company: a ticker (US 'AAPL', TW/JP 4-digit '2330'), a KR 6-digit stock code ('005930'), or a company name in the local language or English ('TSMC', 'Toyota', '삼성전자'). Resolution priority: exact ticker > 6-digit KR code > exact name (name or English name) > partial name (multiple partial matches return a candidate list error). - from / to: optional YYYY-MM-DD range bounds (inclusive) - limit: max rows, 1-500 (default 60) - response_format: 'markdown' (default) or 'json' Account limits: a free account includes the most recent 130 trading sessions of each name. Results follow the current account's history entitlement. When the window is trimmed the response carries a plan_limit field saying so. Returns: {company: {name, source, ticker|stock_code}, count, truncated, prices: [{date, open, high, low, close, volume}]} — newest date first; truncated=true means older rows exist beyond 'limit'. Examples: - {company: '005930', limit: 30} -> last 30 KR trading days for Samsung Electronics - {company: '005930', from: '2026-01-01', to: '2026-06-30'} -> Samsung Electronics H1 2026 Use when: historical closes/volumes for charting or return calculations from ingested data. Don't use for crypto, which is not served. FinBridge has no real-time equity quote tool — equity prices here are end-of-day closes from the nightly ingest; the only live data is regulator filings (get_dart_filings / get_edgar_filings). Errors: unknown company -> no-match or candidate-list error; JP/EU company -> no-prices error (those markets carry statements only); no price rows -> a market-specific hint (new listing, delisted, nightly lag).

Get the latest valuation snapshot for one KR, US, or Taiwan company from the local finbridge database: market cap (latest close x shares) with PER, PBR, PSR, ROE, debt ratio, and 3-year revenue/net-income CAGR, joined to the company's latest annual fundamentals. Includes metric-level calculation basis, dates, sources, missing reasons, and 1-2 same-market percentile hints. Computed by the nightly valuation ingest job. Share counts: KR uses data.go.kr listed shares, US prefers SEC-reported shares, and Taiwan uses exchange-reported shares; the nightly job can fall back to net_income / eps_diluted when a positive result is available. PER prefers price / eps_diluted, falling back to market_cap / net_income. Taiwan exchange-published PER/PBR replace derived values when present. Any derived ratio whose required denominator is null or <= 0 is returned as null. Args: - company: US ticker (e.g. 'AAPL'), KR 6-digit stock code (e.g. '005930'), or company name. Resolution priority: exact ticker > 6-digit code > exact name > partial name (multiple partial matches return a candidate-list error). - per_multiples: optional 1-5 positive user-supplied PER assumptions (maximum 1000) - pbr_multiples: optional 1-5 positive user-supplied PBR assumptions (maximum 100) - response_format: 'markdown' (default) or 'json' Returns the existing valuation fields plus metric_evidence and multiple_scenarios. Ratios are plain numbers; roe/debt_ratio/CAGR and scenario upside/downside are in percent. A scenario is arithmetic from the user's multiple, not a target-price recommendation. If no multiple is supplied, no multiple or target price is invented. Examples: - {company: '005930'} -> Samsung Electronics PER/PBR/ROE plus "PER in the cheapest N% of the KR market" - {company: 'AAPL'} -> Apple valuation snapshot with US-market percentiles Caveats: if statements are in another currency than the listing (Korean listings reporting in USD/CNY/JPY), statement values are converted at one Federal Reserve H.10 rate on or before the price date before PER/PBR/PSR (fx_conversion shows the rate and date; no rate → null). ROE/debt ratio need no conversion; growth across a statement-currency change is null. market_cap keeps the listing currency. KR fundamentals are K-IFRS and US are US-GAAP, so cross-market comparisons are approximate. This is snapshot data (not real-time) and not investment advice. Use when: assessing one company's valuation/quality at a glance, or comparing it to its own market. Don't use for many-company ranking (use query_db) or raw statements (get_dart_financials / get_edgar_financials). Errors: unknown company -> no-match or candidate-list error; 'no valuation snapshot' -> the valuation ingest job has not produced a row for this company (needs a price and latest-annual fundamentals).

Read the companies followed by the authenticated user. Returns names, symbols and markets without changing the watchlist or holdings.