plug.directory
Add my server

Finance Research MCP

by sreenathvemulaco.huggingface.sreenathvemula/financemcp

  • Remote endpoint
  • 61 tools
  • FinanceMCP_ai_chain_profile Where one company sits in the AI value chain of 'US Market Fault Lines' (Links 1-6: lab returns, lab funding, backlog cash, builders' capex and credit, supplier and index earnings, funders' marks) and how exposed it is: role (graph edges and theme segments), capex against operating cash flow and FCF, exposure to the labs (INVESTED_IN rounds, named customer or supplier links with quotes), backlog (RPO) where tagged, financing (maturity ladder, GPU-collateralised 8-K facilities, lenders, hand-collected financing events), share of S&P 500 forward earnings if an AI supplier, disclosed AI or data-centre revenue, the section-10 tripwires that bear on it, and AI hiring. A company the warehouse does not place in the chain is reported as such, with the same cash and exposure sections. Args: ticker: ticker (share classes too), CIK or company name; resolve_company shows how it was read detail: concise (headline values, about 3-4k tokens) | standard (tables) | full (definitions, accession
  • FinanceMCP_ai_cycle The AI-investment cycle against the causal chain in 'US Market Fault Lines' (Links 1-6, section-10 tripwires). view: tripwires (the ten signals computed from the warehouse, with tripped flags and the contradiction test), capex (builders' quarterly capex, OCF, FCF, capex/OCF, RPO; tickers default MSFT, GOOGL, AMZN, META, ORCL, CRWV, NBIS), concentration (AI suppliers' share of S&P 500 forward earnings), private_credit (BDC exposure to software, IT services, AI infrastructure), refinancing (debt ladders of the builders and power names), or all. Args: view: Value for view. Allowed values: ['all', 'tripwires', 'capex', 'concentration', 'private_credit', 'refinancing']. tickers: Value for tickers. quarters: Value for quarters. top_groups: Value for top_groups. detail: concise (headline values, about 3-4k tokens) | standard (tables) | full (definitions, accessions, longer tables) Allowed values: ['concise', 'standard', 'full'].
  • FinanceMCP_business_profile What the company actually DOES: revenue mix (product / geography / segment), operating KPIs (store counts, ARPU, GRM, capacity utilisation, etc.), market share and capex allocation by segment. The starting point for a business/moat/SWOT study. Args: symbol: Value for symbol.
  • FinanceMCP_capital_allocation How management deploys capital over time: operating cash flow vs capex vs free cash flow vs net debt vs dividend payout. Answers 'is growth self-funded or debt-fuelled, and does management return cash' — central to judging stewardship/trustworthiness. Returns a 2D per-year trend chart (PNG) alongside the data when at least 2 years of cash-flow history are available. Args: symbol: Value for symbol.
  • FinanceMCP_company_documents List the filings available RIGHT NOW for a company: concall transcripts and presentations, annual reports, credit-rating rationales and recent announcements, each with a date and a URL. Use this instead of search_documents when you want a company's own words — then call read_document on the ones you need. Works for any listed Indian company without a prebuilt index. Args: symbol: Value for symbol. kind: Value for kind. Allowed values: ['all', 'concalls', 'annual_reports', 'credit_ratings', 'announcements']. limit: Value for limit.
  • FinanceMCP_company_dossier One US-listed company in depth, assembled from the warehouse. sections (comma list; default all): identity, fundamentals (TTM, annual, quarterly, balance sheet, employees), quality (Piotroski, Altman, accruals, ROIC), capital_allocation, growth (CAGRs, base rate for the growth the price implies), valuation (multiples, WACC, reverse DCF), street (consensus targets, ratings, EPS revisions, estimates), guidance (history and guidance versus delivery), governance (CEO pay, pay versus performance, tenure, 13D/G), insiders (open-market buys and sells, recent Form 4 trades), people (officers and directors, leadership changes, H-1B role mix, job postings, layoffs), moat, ip (R&D, patents to 2021, Orange Book), debt (leverage, maturity ladder, recent financings), distress (going-concern and weakness flags), positioning (short interest, borrow, options), events (material 8-Ks), supply_chain (named customers and suppliers at confidence >= min_confidence, disclosed concentration), segments, documen
  • FinanceMCP_company_overview Company profile card: about, sector, key metrics (mcap, PE, ROE/ROCE, sales/PAT), ownership split, index memberships with weights, peers, and available data inventory. Args: symbol: Value for symbol.
  • FinanceMCP_company_valuation What the price implies and what a scenario gives, for one US company. mode='reverse' (default): multiples (P/E, EV/EBIT, EV/EBITDA, EV/sales, P/B, FCF yield), WACC (CAPM with Damodaran implied ERP, own 5y weekly beta) and the constant FCF growth over `years` that the current EV implies, on both FCF bases, with a sensitivity to the discount rate and terminal growth. mode='scenario': two-stage FCF-after-SBC DCF, growth fading linearly from g_start to g_end over growth_years, terminal growth scenario_terminal_growth; every assumption echoed, with a 3x3 sensitivity. mode='all' runs both. discount_rate=0 uses the computed WACC. basis 'reported' (OCF - capex - SBC) or 'maintenance' (OCF - min(D&A, capex) - SBC, for heavy growth-capex years). fcf_base (USD) overrides the TTM base in a scenario when TTM FCF is negative. Outputs are arithmetic on stated inputs, not forecasts or recommendations. Args: ticker: ticker (share classes too), CIK or company name; resolve_company shows how it was
  • FinanceMCP_competitive_position Competitive standing: head-to-head benchmarking vs named peers on operating metrics, market-share trend, and the peer set. Use for 'how does X stack up against competitors' and as the competition input to a SWOT. Args: symbol: Value for symbol.
  • FinanceMCP_country_risk Damodaran country risk for one country: Moody's, S&P and Fitch sovereign ratings, rating-based default spread, 10-year sovereign CDS (and net of the US/Swiss CDS), country risk premium, total equity risk premium (and the CDS-based variants), mature-market and US ERP, corporate marginal tax rate, GDP and PRS score - from the latest update held (July 2026 at present) and the April 2026 update, plus a history table across every update in the warehouse (January 2026, July 2025, ...) and the latest monthly implied ERP of the S&P 500. country: a name ('United States', 'Taiwan', 'India', 'Germany'; US/UK/UAE accepted) or '?' to list them. Args: country: country name, or '?' to list detail: concise (headline values) | standard (every curated metric / longer tables) | full (every column of every file; get_series up to 5,000 points) Allowed values: ['concise', 'standard', 'full'].
  • FinanceMCP_customers_of Named customers of a company with verbatim quotes, source, date, confidence, products and revenue share where stated (same fields as suppliers_of). Anonymised customers ('Customer A accounted for 22%') are in doc_facts(fact_type='customer_concentration'); federal-government customers appear with node org:us-federal-government when an XBRL fact or sentence names them. Args: ticker: ticker (share classes too), CIK or company name; resolve_company shows how it was read min_confidence: Value for min_confidence. limit: Value for limit. detail: Value for detail. Allowed values: ['concise', 'full']. include_unchecked: Value for include_unchecked.
  • FinanceMCP_demand_signals AI demand signals: quote-backed statements from the earnings calls (last 8) and 10-Q MD&A (last 6) of the AI-chain companies about customer orders, backlog, capex plans, supply constraints and inventory, each with company, speaker, customer type (hyperscaler, neocloud, enterprise, sovereign, china, consumer), named customer, product, direction, stated magnitude and period, the verbatim quote, date and chunk_id (doc_read(chunk_id) shows the surrounding text). Use it for 'are customers cutting or pausing AI spend', 'who reports pull-ins or push-outs' before searching raw text. ticker: one or a comma list. theme: ai-chain node (labs_and_apps, cloud_builders, data_centers, accelerators, memory, foundry_packaging, equipment, networking_optics, servers, power, fuel) or 'negative' (decrease, pause_digestion, push_out, cancellation). direction: increase, decrease, pause_digestion, push_out, pull_in, cancellation, steady or 'negative'. signal_type: customer_demand, own_capex, supply_constraint,
  • FinanceMCP_doc_facts Facts read out of company text, each with its sentence, source and (where the passage is in the search index) the chunk_id to open with doc_read. fact_type: guidance (forward guidance from 8-K releases and calls: metric, period, low/high/mid, unit, basis; plus revenue guidance against what was then reported), capex_plan (capex guidance only), tam_claim (market-size claims by the company: issuer claims, not measurements), customer_concentration (customer and supplier concentration disclosed in filings, including anonymised customers). metric filters guidance or TAM rows by text; since is YYYY-MM-DD. Not extracted as tables: headcount, segment changes, risk-factor changes, litigation (the error says how to find them with doc_search). Args: ticker: ticker (share classes too), CIK or company name; resolve_company shows how it was read fact_type: Value for fact_type. Allowed values: ['guidance', 'capex_plan', 'tam_claim', 'customer_concentration']. metric: Value for metric.
  • FinanceMCP_doc_read Read the stored text behind a chunk_id from doc_search, doc_facts, demand_signals or a supplier/customer quote. before / after (0 to 3) add the neighbouring chunks of the same document item; max_chars (500 to 20,000, default 6,000) caps the text. Returns ticker, company, document, date, section or speaker, the source URL and the text. Read before you quote. Args: chunk_id: Value for chunk_id. before: Value for before. after: Value for after. max_chars: Value for max_chars.
  • FinanceMCP_doc_search Search the filings and earnings calls of the S&P 500 + AI chain by meaning and keywords and get about 40 CANDIDATE passages (10-K: last 3 fiscal years; 10-Q: last 6; calls: last 12; proxy statements DEF 14A, 8-K earnings releases and material agreements; foreign private issuers' 20-F and 6-K). Each candidate: chunk_id, ticker, document type and date, fiscal period, section or speaker (management / analyst), and an excerpt of about 320 characters chosen for the query. The candidates are NOT ranked by relevance (BM25 and dense retrieval fused by rank, no reranker): YOU judge them, open the promising ones with doc_read(chunk_id), and quote only from what you have read. When the results are weak, reformulate: use the company's own terms, name the period, add tickers / doc_types / items filters. Ask the question in plain words: a named company becomes a filter, 'June 2026 call', 'Q2 2026', 'latest', 'what changed in the 10-K' narrow or prefer the right documents (the header line says which
  • FinanceMCP_financial_health Multi-year fundamental HEALTH check (~12y): computes sales/profit/EPS CAGR, margin trend, earnings quality (cumulative operating-cash-flow vs profit = accruals check), interest coverage, debt trend, genuine per-share dilution, ROCE trend and debtor-day creep — then emits directional flags (concern / watch / strength). Auto-adjusts for banks/NBFCs (skips cash-conversion & working-capital checks that don't apply to lenders). This is the go-to tool for 'find issues with the financials'. Args: symbol: Value for symbol. basis: default: consolidated, fallback standalone Allowed values: ['consolidated', 'standalone'].
  • FinanceMCP_financial_statements Historical financial statements: Indian companies from screener data (annual, ~10-12 years, Rs crore), US tickers from SEC EDGAR (annual or quarterly, USD millions, EPS in USD per share; the output states the units). statement is one of profit_loss | balance_sheet | cash_flow | quarterly_results | ratios | shareholding. 'ratios' returns the efficiency series (debtor days, inventory days, cash conversion cycle, ROCE) that the other statements do not carry (Indian companies only). For US tickers the profit_loss table carries EPS twice: 'as filed' (each year on the share count of the filing that last reported it, so it jumps across stock splits) and 'split-adjusted' (today's share count) — use the adjusted row for growth rates. Args: symbol: Value for symbol. statement: Value for statement. Allowed values: ['balance_sheet', 'cash_flow', 'profit_loss', 'quarterly_results', 'ratios', 'shareholding'].
  • FinanceMCP_find_data Find what the warehouse holds: tables/views (name, rows, which tool already serves it), time series (series_id, name, units, frequency) and registered sources (publisher, licence, frequency, last fetch, causal links). All words in `text` must match. kind: '' (all three), 'table', 'series', 'source'. source restricts series to one source id (e.g. 'eia_pet'); link restricts sources to one causal link (link1..link6, mechA, mechB, rates, oil, power, china, valuation). Lists names and descriptions only; the values are served by the tool named in served_by (company, market and document tools; industry_dashboard, industry_compare and country_risk for the Damodaran tables; search_series and get_series for time series). Args: text: Value for text. kind: Value for kind. Allowed values: ['', 'table', 'series', 'source']. limit: Value for limit. detail: concise (headline values, about 3-4k tokens) | standard (tables) | full (definitions, accessions, longer tables) Allowed values:
  • FinanceMCP_forensic_checks Forensic / governance red-flag checklist (from tijori): ~17 plain-English checks grouped by topic (Accounting Quality, Promoter & Governance, Balance Sheet & Debt, Growth & Returns, Valuation & Sentiment) plus an explicit promoter-pledge reading. Use alongside financial_health when scrutinising accounting quality or management integrity. Args: symbol: Value for symbol.
  • FinanceMCP_get_series Values of one warehouse series with its name, units, frequency, source and last date, and simple statistics over the full history: latest value, 1-year and 3-year % and absolute change (against the nearest earlier observation), percentile of the latest value in its own history, median, min, max. Points are returned as 'period,value' lines: at most 500 (the latest in the window) by default and up to 5,000 with detail='full'. start / end: YYYY, YYYY-MM or YYYY-MM-DD. freq: '' (native), D, W, M, Q or A with agg last (default), mean or sum. source: only when the same id exists in two sources ('source:series_id' also works). The statistics use observations up to today (EIA STEO and AEO forecasts are returned as points but not used for latest or changes). DRAMeXchange history starts with the first page snapshot (October 2026). Only series held in the warehouse (search_series finds the id); no SQL; India-only sources are refused. Args: series_id: series id from search_series or find_data
  • FinanceMCP_guidance_history Forward guidance from 8-K earnings releases and earnings calls. For the AI chain and the 100 largest S&P names it reads guidance_verified (each sentence read and structured: metric, segment, period, low/high/point/mid, unit, GAAP basis, direction vs prior; non-guidance removed). Other tickers fall back to the rule-extracted table (about 8% of its rows have an error). Read the sentence before relying on a figure. doc_facts(fact_type='guidance') adds guidance against delivery and the chunk_id of each sentence. Args: ticker: ticker (share classes too), CIK or company name; resolve_company shows how it was read metric: filter by metric text, e.g. capex, revenue, gross margin
  • FinanceMCP_index_data NSE index data. kind=prices: OHLCV history (monthly). kind=valuation: PE/PB/div-yield history. kind=list: available indices. Names e.g. nifty50, nifty500, bank, it, auto, energy, fmcg, metal, pharma. Args: index: Value for index. kind: Value for kind. Allowed values: ['prices', 'valuation', 'list'].
  • FinanceMCP_industry_compare Several Damodaran industries side by side (up to 12, plus Total Market for reference): each metric's value and the industry's percentile among the 94 industries (pNN, 100 = highest). industries: names separated by ';' (tickers are mapped through the SIC mapping), e.g. 'Semiconductor; Power; Utility (General); NVDA'. metrics: comma list of keys (rev_cagr_5y, exp_rev_growth_5y, exp_eps_growth_5y, gross_margin, op_margin, ebitda_margin, net_margin, sbc_sales, roe, roic, cost_of_capital, cost_of_equity, roc_minus_wacc, roe_minus_coe, roic_minus_coc, beta, unlevered_beta_cash, pe_trailing, pe_forward, peg, pbv, ev_sales, ev_ebitda, capex_da, net_capex_sales, sales_invested_capital, nwc_sales, market_de, debt_ebitda, interest_coverage, payout, dividend_yield, cash_return_ni, rnd_sales, n_firms ...), a group (growth, margins, returns, cost_of_capital, risk, multiples, reinvestment, working_capital, debt, payout, rnd, size, ownership_tax) or 'all'; default: a 17-metric overview. region: US, Gl
  • FinanceMCP_industry_dashboard Damodaran industry data for one industry: every metric of his industry-average files with its percentile among the 94 industries and the dataset date - growth (5y revenue and net income CAGR, analysts' expected revenue and EPS growth, fundamental growth = ROE x retention and ROC x reinvestment), margins (gross, operating, lease and R&D adjusted, EBITDA, net, SBC/sales), returns (ROE, ROIC), cost of capital (cost of equity, after-tax cost of debt, D/(D+E)) and the excess return (ROC - WACC, ROE - COE), betas (levered, unlevered, cash-corrected), multiples (P/E trailing and forward, PEG, P/B, EV/sales, EV/EBITDA), reinvestment (capex/D&A, net capex/sales, sales/invested capital), working capital, debt (D/E, debt/EBITDA, interest cover), dividends and buybacks, R&D/sales, number of firms. Dataset: Damodaran's January 2026 update (dated 2026-01-05), US and global files. industry: a Damodaran name ('Semiconductor', 'Software (System & Application)', 'Power', 'Utility (General)') or '?' to l
  • FinanceMCP_inputs_of Inputs, raw materials and key dependencies of one AI-chain company (58 US-listed and SEC-filing names, 54 of them with text in the document index: accelerators, memory, foundry, equipment, networking, servers, data centres, cloud builders, labs/apps, power, fuel), each quoted from its 10-K/10-Q/20-F or calls: category (raw material, component, capacity, energy, service), named suppliers, single-source and country risk, and disclosed cost information with its period (amounts, purchase obligations, raw-material inventory, price-change statements). Company totals (purchase obligations, raw-material inventory) come first. Each input is mapped to a public market price series computed live from the warehouse (latest value and date, 1-year and 3-year % change), flagged when the series is a proxy, when no public price exists, or when a free series is not yet in the warehouse. Companies rarely disclose per-material costs, and market prices show trends, not the company's contract prices. detail=
  • FinanceMCP_insider_trading PIT (insider trading) disclosures for a company, if available. Args: symbol: Value for symbol.
  • FinanceMCP_kg_entity Resolve a company, fund, person, lender, AI lab, theme or sector to knowledge-graph node ids. Accepts a ticker (GOOG and GOOGL are one node), CIK, node id (cik:1045810, fund:2012383, lab:openai, lender:mufg-bank, sector:3674, theme:ai-link-5), or a name (case, punctuation and Inc/Corp/Ltd/Holdings/Group/The are ignored; all-words and fuzzy fall back). Returns the match method, graph degree and same_as ids. Use it before kg_neighbors when a name could mean several nodes. Args: name_or_ticker: Value for name_or_ticker. limit: Value for limit.
  • FinanceMCP_kg_neighbors Relationships around an entity in the knowledge graph: counts per relation, then the top rows spread across relations, each with value, detail, date, source table and key and confidence. Relations: OFFICER_OF, DIRECTOR_OF, INSIDER_OF (Form 3/4/5), PAID (exec pay), HOLDS_13F (latest-quarter fund positions), HOLDS_13DG (>5% holders), INVESTED_IN (investor -> AI lab), LENDS_TO / UNDERWRITES / TRUSTEE_OF (8-K debt terms), IN_THEME / PART_OF / AI_CHAIN_LINK, IN_SECTOR (SIC), FILES_H1B, HAS_PATENTS, AWARDED (CHIPS Act), SAME_AS, AFFILIATE_OF; text-extracted CUSTOMER_OF, SUPPLIER_OF, COMPETES_WITH, PARTNERS_WITH carry an evidence quote. '->' = the entity is the source of the edge, '<-' = the target. rels: comma list to filter. depth=2 adds neighbours of the strongest first-hop nodes. as_of: YYYY-MM-DD for a point-in-time view (default current). Confidence below 1 means a name match or an inferred direction: read it before relying on an edge. A name that matches only partly is not used: pass a
  • FinanceMCP_kg_path Shortest connections between two entities in the knowledge graph (edges followed in either direction; a fund and its listed company are one entity). Direct edges come first. Otherwise the search runs in tiers and stops at the first that connects them: (1) people, investors, lenders, text-extracted customers/suppliers/partners/competitors; (2) plus 5% holders, programme awards, themes, AI-chain links; (3) plus 13F index-fund co-holding. The output says which tier was needed, because tiers 2 and 3 are weak links. Examples: kg_path('NVDA','CoreWeave'), kg_path('Microsoft','Anthropic'). Args: a: Value for a. b: Value for b. max_depth: Value for max_depth. rels: Value for rels. as_of: Value for as_of. min_confidence: Value for min_confidence. k: Value for k.
  • FinanceMCP_kg_shared What two entities have in common in the knowledge graph: shared directors and officers, 13F and 13D/G holders (each side's value or % of class), lenders and arrangers, investors, common investments, themes, sector, AI-chain links, H-1B/patent hubs and text-extracted customers, suppliers, partners and competitors, plus any direct edge between the two. Counts per category, then top rows with each side's largest weight, date and source. Args: a: Value for a. b: Value for b. limit: Value for limit. rels: Value for rels. as_of: Value for as_of. min_confidence: Value for min_confidence.
  • FinanceMCP_macro_data Indian macro series (monthly/annual CSVs): cpi_yoy, gsec10y, iip, policy_rate, rate3m, usdinr, macro_monthly, macro_annual. Omit 'series' to list every series the dataset ships, plus live market proxies (crude, gold, silver, and us10y = the US 10-year Treasury yield, which is not an Indian rate). Returns the most recent 24 observations; the last row's date is the series' own as-of date. Args: series: Value for series.
  • FinanceMCP_management_guidance MANAGEMENT CREDIBILITY tracker: retrieves past forward-looking / guidance statements from earnings calls (revenue/margin/capex/outlook) and pairs them with the ACTUAL quarterly results that followed, so you can judge whether management delivered on its promises. Surfaces both sides as evidence — you make the trustworthiness call. Requires the document index; returns actuals + a notice if no concalls are indexed. Args: symbol: Value for symbol. lookback_periods: Value for lookback_periods.
  • FinanceMCP_market_dashboard US market as a whole, each with today's value and its percentile in its own history: Shiller CAPE (since 1881 and 1990), FactSet forward P/E, forward earnings yield minus real 10y, Damodaran implied ERP, nonfinancial corporate equity/GDP, profits/GDP, HY OAS, Baa-10y, NFCI, St Louis stress, curve, 10y nominal/real, 5y5y inflation, Sahm rule, margin debt, S&P 500 concentration, AI suppliers' share of index NTM earnings; plus OFR stress index, excess bond premium and its recession probability, Tobin's q, household equity share of financial assets, nonfinancial corporate debt/GDP, BIS credit gap, ACM 10y term premium, DIX/GEX, CFTC leveraged-fund net E-mini S&P and Nasdaq, and the Treasury General Account. Args:
  • FinanceMCP_material_exposure Which AI-chain companies depend on one material or input (e.g. copper, aluminium, HBM, DRAM, NAND, substrates, CoWoS/advanced packaging, foundry wafers, GPUs, electricity, natural gas, uranium/nuclear fuel, gallium, germanium, rare earths, helium/gases, transformers, data-centre space), with each company's disclosed dependence, supplier, single-source/country risk and cost notes (quote-backed, with period), and the material's market price trend computed live from the warehouse (best series and proxies: latest value and date, 1-year and 3-year % change; free series not yet in the warehouse are named). Companies rarely disclose per-material costs; market prices show trends, not contract prices. detail='full' adds the verbatim quotes and chunk_ids. Args: material: material key or name, e.g. copper, hbm, dram, natural gas, electricity, uranium detail: Value for detail. Allowed values: ['concise', 'full'].
  • FinanceMCP_peer_compare Compare a company with its peers, or any list of tickers, on a preset: profits (margins), cash (FCF margin, cash conversion, capex, SBC, dilution), growth (CAGRs, percentile in its SIC group), returns (ROIC, ROIC minus industry cost of capital, ROE), balance_sheet (net debt/EBITDA, interest cover, maturities), valuation (EV multiples, P/E, FCF yield, reverse-DCF growth), people (revenue per employee where tagged, H-1B role mix, open roles, CEO pay ratio, TSR vs peers), management (guidance hit rate and bias, insider net buying, 13D/G filings, CEO tenure), moat, ip, or all (comma list allowed). Give `ticker` to compare with its peer_set (found by 2+ sources), and/or `tickers` (comma list) for an explicit set. Each metric row has every ticker's value, the peer median and quartiles (ratios only), and the focus company's rank and percentile; fiscal-year-end mismatches are flagged; missing values carry reasons. period 'fy' (latest fiscal year) or 'med5' (median of five fiscal years). indust
  • FinanceMCP_peer_set Peers of one company, with the source of each: same 4-digit SIC, same Damodaran industry (via the tool's SIC mapping), COMPETES_WITH edges read from 10-K text, Hoberg-Phillips TNIC3 text similarity (to 2023), shared theme segments. A peer found by several sources ranks first; ties go to the closer revenue size. Give ticker for peers of a company, or sector for the members of a sector: a SIC prefix ('3674'), a theme or segment ('ai', 'ai.accelerators', 'energy'), or a Damodaran industry name ('Semiconductor'). Feeds peer_compare. Args: ticker: Value for ticker. sector: Value for sector. n: Value for n. detail: concise (headline values, about 3-4k tokens) | standard (tables) | full (definitions, accessions, longer tables) Allowed values: ['concise', 'standard', 'full'].
  • FinanceMCP_peers_and_index Peer companies (same industry) and index memberships for a symbol. Args: symbol: Value for symbol.
  • FinanceMCP_portfolio_risk Portfolio-level risk from ACTUAL historical daily returns — not single-stock volatility/beta treated as a portfolio proxy. Takes named holdings + weights (any positive numbers; renormalized to sum to 1, flagged if so), pulls each symbol's real daily price history, and computes: the correlation matrix between holdings, portfolio annualized volatility (Markowitz w'*cov*w, cross-checked two ways), and — via the `empyrical` risk-stats library (empyrical-reloaded, the maintained fork of Quantopian's widely-used open-source package) applied to the actual weighted portfolio-return series — CAGR/max drawdown/Calmar ratio, Sharpe and Sortino ratios (risk-free rate from macro_data's 10y G-Sec unless supplied), HISTORICAL (empirical-percentile, non-parametric) VaR and CVaR at 95%/99% — not a normal-distribution z-score approximation, and portfolio beta vs a benchmark index cross-checked against the weighted average of individual betas. Also reports a Herfindahl-index-based 'effective number of po
  • FinanceMCP_price_analytics Trader-oriented price statistics beyond basic momentum: 52-week range & position, distance from all-time high, max drawdown & current drawdown, annualised volatility, volume trend vs 200-day average, 50/200-DMA crossover (golden/death cross), and relative strength vs NIFTY over 3m/1y. Performance metrics use split/bonus-adjusted prices. Args: symbol: Value for symbol. benchmark: index for relative strength (nifty50, it, bank, ...)
  • FinanceMCP_price_history OHLCV price history between dates (YYYY-MM-DD). interval: daily|weekly|monthly. Use for 'what happened to the stock around <date>' style questions. Args: symbol: Value for symbol. start: Value for start. end: Value for end. interval: Value for interval. Allowed values: ['daily', 'monthly', 'weekly'].
  • FinanceMCP_read_document Return the text of a filing (concall transcript, annual report, rating rationale, announcement) listed by company_documents. ALWAYS pass symbol as well as url: the archive already holds the parsed text of most listed filings, which is cleaner than a fresh PDF extraction and works even when the exchange refuses the download from this server's region. Falls back to downloading the PDF when the archive does not have it. SEC filings (an https://www.sec.gov/Archives/... HTML URL from us_executive_pay or the US filings lists: 10-K, 10-Q, DEF 14A) are fetched from EDGAR and converted to text. Long documents come back in windows of max_chars (default 40,000) starting at offset; the last line says which offset continues the text. A transcript is typically 15-20 pages / ~12k tokens, so read it directly rather than searching an index. Quote it with the document type and period. Args: url: URL from company_documents symbol: the company the URL came from. Pass it: the archive usually alrea
  • FinanceMCP_refresh_us_data Rebuild the US screening table (used by screen_stocks market='us' and the US columns elsewhere) from the SEC's bulk Company Facts archive, which the SEC regenerates nightly. US ONLY -- Indian data is not touched by this tool. action='status' (default) says when the table was built, which SEC archive it came from, whether that archive has changed since, and the progress of any refresh under way. action='start' begins a refresh in the background and returns at once: the download is ~1.4 GB and parsing ~20,000 filers takes several minutes, so call action='status' again later rather than waiting. If the SEC archive is unchanged since the last build nothing is downloaded (force=true overrides that). Refreshes are limited to one per ~20 hours. The refreshed table is used immediately; whether it also survives a server restart depends on the server's token having write access to the dataset, and the status says which. When the server is protected, action='start' needs the owner's passphrase --
  • FinanceMCP_resolve_company Start here for any company. Resolves a ticker, share-class ticker (GOOG), CIK, predecessor CIK or name. market='auto' (default): a US-listed company returns the canonical record (ticker, CIK, name, SIC, fiscal year end, share classes, predecessor CIKs, knowledge-graph node id for kg_* tools, Damodaran industry, and which data families the warehouse holds with dates); a name that is not an exact US ticker or CIK also searches the Indian NSE/BSE universe and lists those symbols. market='us' or 'india' restricts it. `query` is the older argument name of the Indian-only tool: a call with `query` and no market searches NSE/BSE only (an Indian symbol such as TCS is also a US ticker). Args: text: ticker, CIK or company name query: older argument name of the Indian tool; with query and no market only NSE/BSE is searched detail: concise (headline values, about 3-4k tokens) | standard (tables) | full (definitions, accessions, longer tables) Allowed values: ['concise', 'standard', 'f
  • FinanceMCP_screen_by_year Screen the universe by a SPECIFIC HISTORICAL YEAR, not today's snapshot — answers 'ROCE > 20% in FY2024' or 'best/worst price performers in 2023'. kind='fundamental' (default): year = fiscal year ending March (year=2024 means FY ending Mar 2024); filter/sort columns: sales_cr, opm_pct, net_profit_cr, eps, roce_pct (non-financials), roe_pct (banks/NBFCs), sales_yoy_growth_pct, net_profit_yoy_growth_pct, eps_yoy_growth_pct. kind='price_return': year = calendar year; filter/sort on return_pct (split/bonus/dividend-adjusted close-to-close for that year); it reads a precomputed table and reports 'not available on this deployment' where that table is not shipped. This is a point-in-time screen — not a recommendation, and extreme ratios on tiny-capital-base companies should be sanity-checked before acting on them. Args: year: fiscal year (Mar-end) or calendar year per 'kind' kind: Value for kind. Allowed values: ['fundamental', 'price_return']. min: Value for min. max: Value
  • FinanceMCP_screen_consistency Screen the WHOLE universe for names that clear a bar EVERY year over N years — not just the latest year (screen_by_year) or today's snapshot (screen_stocks). This is the Coffee Can-style 'ROCE>=15% every year for 10 years' rule in one call, done for every company at once, instead of a coarse cut plus manual per-candidate verification. Pass any of: sales_cr, opm_pct, net_profit_cr, eps, roce_pct, roe_pct, sales_yoy_growth_pct, net_profit_yoy_growth_pct, eps_yoy_growth_pct — or metric='roce_or_roe' for a mixed universe (this is NOT a separate metric, it just applies ROCE to non-financials and ROE to banks/NBFCs per company, the standard Coffee Can convention — the output's metric_used column always states which one was actually used per row). max_violations lets a small number of off years through (0 = strict, every single year). Companies with less history than min_years_required are excluded, not penalised, if newer than the window. To combine two rules (e.g. Coffee Can's ROCE/ROE AND
  • FinanceMCP_screen_stocks Objective stock screen over ~3100 NSE/BSE companies, ~45 filterable numeric columns. Filter by sector/industry text, index membership, and numeric minimums/maximums on: VALUATION: market_cap_cr, price, pe, pb, ps, ev_ebitda, div_yield, earnings_yield_pct (100/PE), fcf_yield_pct (fcf_cr / market cap), book_value_per_share (price / PB), peg (PE / 3-year EPS CAGR; undefined unless both are positive). PROFITABILITY/RETURNS: roe, roce, roic_pct, roa_pct, gross_margin_pct, opm_pct (latest fiscal year), net_margin_pct. GROWTH (latest fiscal year, from the shipped annual table; the growth_fy column says which year): sales_growth_pct, eps_growth_pct (YoY), eps_cagr_3y_pct. LEVERAGE/LIQUIDITY/EFFICIENCY: debt_equity, interest_coverage_x, current_ratio, quick_ratio, asset_turnover, fixed_asset_turnover, equity_multiplier, fcf_margin_pct, fcf_cr, net_debt_cr, total_debt_cr, debtor_days, inventory_days, payable_days, cash_conversion_cycle. GOVERNANCE: promoter_pledge_pct. (Promoter/FII/DII/public s
  • FinanceMCP_search_documents Semantic search over the indexed document corpus: earnings-call transcripts (Q&A level) and presentations, annual reports, credit-rating rationales, corporate announcements, XBRL filings. Filter by symbol, doc_types, and date range (YYYYMMDD ints — note: undated docs like announcements are excluded when date filters are set). For cross-company thematic queries (no symbol), set max_per_symbol (e.g. 2) so one company doesn't dominate. Use for qualitative questions: guidance, strategy, risks, capex plans, management commentary, rating rationale, segment detail. For NUMBERS use financial_statements instead. Args: query: Value for query. symbol: Value for symbol. doc_types: Value for doc_types. date_from: e.g. 20240101 date_to: Value for date_to. max_per_symbol: cap hits per company on cross-company queries top_k: Value for top_k.
  • FinanceMCP_search_series Find time series in the warehouse (1.18 million series__ series: FRED, ICE BofA spreads, EIA power, gas, oil and outlook, World Bank commodity prices, BLS, BEA, Census, Fed, Treasury, Taiwan AI-chain revenue, ...; plus derived series: DRAMeXchange memory spot prices, GPU rents (monthly index, getdeploying, GPU Economy), OpenRouter token counts and China-lab share, Taiwan MOPS monthly revenue of every listed company). All words must appear in the series id, name or the source description (as substrings, so short words such as 'tic' match widely); results are ranked (phrase and word-start matches in the name, headline sources first) with the count per source. source narrows to one source id (e.g. 'eia_ng', 'fred', 'dramexchange', 'gpu_rent', 'openrouter', 'tw_mops_revenue'). India-only sources are not served. get_series(series_id) returns the values. Args: text: words, e.g. 'henry hub', 'copper', 'DRAM', 'Taiwan revenue', 'H100' source: Value for source. limit: Value for lim
  • FinanceMCP_sector_analysis Aggregate view of a sector or industry: company count, total & median market cap, median valuation, and the leaders by market cap with headline metrics. Pass sector OR industry text (partial match, e.g. 'Banks', 'IT - Software', 'Pharmaceuticals'). Omit both to list available sectors. Use for 'analyse the X sector' instead of looping companies. Args: sector: Value for sector. industry: Value for industry. top_n: Value for top_n.
  • FinanceMCP_shareholding_trends Quarter-over-quarter ownership shifts: promoter stake direction, FII/DII flows and promoter share pledge. Declining promoter stake and/or rising pledge is a classic governance warning. Complements the shareholding snapshot in financial_statements. Args: symbol: Value for symbol.
  • FinanceMCP_suppliers_of Named suppliers of a company (foundries, memory, contract manufacturers, cloud and software vendors ...) from 10-Ks, notes, 8-Ks and earnings calls, each with a verbatim quote, its source and date, a heuristic confidence (higher when both parties name each other or the source is a 10-K / XBRL fact), products and revenue share where stated. ticker accepts a ticker, CIK, node id (cik:1045810) or name. Coverage is what filings and calls name: absence is not evidence of no relationship. detail='full' gives the whole sentence and URL. Rows from the manual review (extractor claude-review) come first; extractor rows with the wrong direction, a litigation or trade-restriction sentence, or no rule or structured fact behind them are hidden and counted (include_unchecked=true shows them). Args: ticker: ticker (share classes too), CIK or company name; resolve_company shows how it was read min_confidence: Value for min_confidence. limit: Value for limit. detail: Value for detail. A
  • FinanceMCP_supply_chain Supply chain of a company. US-listed (ticker, CIK, node id or name): upstream (suppliers, then their suppliers) and downstream (customers, then their customers), depth 1 to 3, following the strongest per_node links from each company; each link is 'A supplies B' with confidence, products, source and a short quote, built from named relationships only. Indian company (NSE/BSE `symbol`): the older tool, procurement-cost drivers, named customers and what needs web research. market='auto' (default) uses `ticker` for the US and `symbol` for India; if a ticker is unknown in the US warehouse it is tried as an Indian symbol. Args: ticker: US ticker, CIK, node id or name symbol: NSE/BSE symbol (Indian companies) depth: Value for depth. min_confidence: Value for min_confidence. per_node: Value for per_node. include_unchecked: US only: also show edges the quality screen hides (wrong direction, litigation context, no rule behind them) market: auto (default) | us | india
  • FinanceMCP_technicals_momentum Technical indicators & momentum: price vs 50/200-DMA, RSI, MACD, returns over 1d/1w/1m/3m/6m/1y/3y, 52-week-high distance. live=true recomputes from the price history parquet (freshest); live=false reads the stored snapshot. Args: symbol: Value for symbol. live: Value for live.
  • FinanceMCP_topic_timeline How a topic EVOLVED across quarters for one company: retrieves the best chunk(s) from each of the last N periods separately (one search per quarter, newest first), so every period is represented. Use for questions like 'how did margin guidance change over the last 2 years' or 'track capex commentary quarter by quarter'. Args: query: Value for query. symbol: Value for symbol. doc_types: default: concall transcripts + presentations n_periods: Value for n_periods. per_period: Value for per_period.
  • FinanceMCP_us_earnings_call US earnings-call transcripts (the US counterpart of the Indian concall transcripts). Coverage: the kurry/sp500_earnings_transcripts dataset (MIT), about 33,000 calls for 685 S&P 500 companies, 2005-2025, with the dataset's own year/quarter labels; companies outside it, and calls after 2025, are absent. With only symbol (and optionally year) it LISTS the available calls with their dates. With year and quarter (or latest=true) it returns that call's speaker-tagged text. section: full | prepared_remarks | qa — the Q&A is taken to start at the first operator turn after the opening, which the output states; where a call has none, prepared_remarks/qa are refused and full still works. Long calls are returned in windows of max_chars (default 40,000) from offset; the last line says which offset continues. The first call after a restart downloads a 1.8 GB data file and can take a minute or more; later calls are fast. Args: symbol: US ticker, e.g. NVDA year: the dataset's year label for
  • FinanceMCP_us_executive_pay Locate executive-compensation disclosure for a US company: the DEF 14A proxy statements containing the Summary Compensation Table (salary, bonus, stock and option awards per named officer), CEO pay ratio, pay-versus-performance, and the beneficial-ownership table. Returns filing URLs to read with read_document — the figures are deliberately not auto-parsed, because compensation tables vary in shape between filers and mis-attributing pay between officers is worse than reading the table. Args: symbol: Value for symbol.
  • FinanceMCP_us_insider_trades US insider dealing from SEC Forms 3/4/5 for one company and quarter. Officers and directors must report within two business days, so this is close to real time. Separates open-market BUYS and SELLS (codes P/S — actual decisions) from grants, option exercises and tax withholding (A/M/F — compensation mechanics), which must not be read as conviction trades. Covers every SEC filer. Args: symbol: Value for symbol. year: calendar year, e.g. 2026 quarter: Value for quarter. Allowed values: [1, 2, 3, 4]. limit: Value for limit.
  • FinanceMCP_us_screen Screen currently listed US SEC filers on their latest fiscal year (full XBRL panel): 5y revenue CAGR, gross/EBIT/FCF margins, ROIC, net debt/EBIT, SBC % revenue, 3y diluted-share CAGR. Ratios are decimals (0.15 = 15%); the defaults (-9, 99, 9) mean no filter. sic_prefix e.g. '3674' semis, '737' software, '49' utilities. Returns the funnel (names left after each filter), the passing names (at most `limit`, 100 max), the NEAR MISSES (fail exactly one filter, with the filter that excluded them) and a percentile composite when rank_by='composite' (mean percentile of `composite` metrics, comma list among roic, op_margin, fcf_margin, rev_cagr_5y, gross_margin, sbc_pct_revenue, share_cagr_3y, net_debt_to_ebit; lower-is-better metrics inverted). rank_by or sort_by: revenue_bn (default), composite, or any output column. Banks, insurers and REITs have no meaningful gross margin or ROIC here: exclude them with a SIC prefix. Args: sic_prefix: Value for sic_prefix. min_revenue_bn: Value fo
  • FinanceMCP_us_status State of the US warehouse and document-search runtime: whether the search index is loaded, which build, chunk count, memory, tables fetched, last refresh check. Use it when doc_search says the index is loading. Args:
  • FinanceMCP_valuation_summary Valuation workup for a company: current multiples (PE/PB/PS/EV-EBITDA/yield, ROE, ROCE, D/E), relative valuation vs peers/sector-index/own 10y history, and a 3-scenario DCF (bear/base/bull with implied growth). Includes the data-source disclaimer. Args: symbol: Value for symbol.
  • FinanceMCP_xbrl_quarterly NSE XBRL quarterly filings, straight from the exchange filing rather than a data vendor: ~150-200 line items per quarter (full P&L, balance sheet, cash flow, EPS, paid-up capital, audited/unaudited flag), Rs crore, up to ~26 quarters, for ~2,000 companies. Use when financial_statements(quarterly_results) is too coarse -- it carries ~12 rows; this carries every filed line. Flow items are returned as SINGLE QUARTERS: NSE half-year (Sep) and year-end (Mar) filings are year-to-date, so those columns are de-cumulated by subtraction and marked *; verified against screener quarterly figures (net profit matches to the rupee on TCS/INFY/3MINDIA/RELIANCE). A cell tagged (H1)/(FY) is still year-to-date because the quarters needed were not on file. Revenue can differ from screener 'Sales' by a definitional gap (gross vs net of excise/GST, e.g. ITC +8-14%) -- that is the filing, not an error. Default output is a headline block; pass line_items=['Borrowings', 'Trade Receivables'] to pull any row by

As the server's own listing gives them. We never call a tool.