ai.jawz/jawz

Measured conclusion: Deep-tested. L2 outcome: tools_ok.

Live macro data for AI agents: regime, conditions, liquidity, with input provenance. No signup.

Add to your agent

Copy this into your MCP client config:

{
  "mcpServers": {
    "jawz": {
      "url": "https://jawz.ai/api/mcp",
      "transport": "streamable-http"
    }
  }
}

Measured tools (24)

• get_started read-only

  Orientation for a new Jawz connection, and for questions about what Jawz is, what it covers, or how it is used. Returns where to start (start_here: get_dashboard, the Jawz Dashboard, which gives the state of the world in six readings), a description of the Jawz Loop, example opening prompts, a one-line live market read drawn from the current data layer, a note on the plain-English glossary included in Jawz responses, and a link to the full guide.

  Hints: read-only yes · destructive no · idempotent not declared · open-world no

• get_dashboard read-only

  The Jawz Dashboard: where the world stands, in six readings. Regime (GREEN / YELLOW / RED: consumer mood and money together), Money (PLENTIFUL / STEADY / SCARCER: the Fed, ECB, Bank of Japan and PBoC balance sheets in dollars), Business cycle (SPRING / SUMMER / FALL / WINTER: growth, a vote of factory output, jobs and investment, crossed with prices), Financial conditions (LOOSE / EASING / NEUTRAL / TIGHTENING / TIGHT: rates and credit), Store of value (RISING / MIXED / FALLING: gold and bitcoin) and Investment (BOOMING / STEADY / FADING: the AI build-out, from data centres, chips, the four big builders' capital spending and power). Each card carries its rule in one sentence, the badge with the date it has held since, the badge's last three prints, and the rows behind it, each with its last three prints (date, word and value as the card writes it) and a data-quality status (ok, stale or unavailable). The summary has the six badges, what set each, and the biggest risk in one line. Returned three ways: the dashboard object as structured content, a short card as markdown text (one line per reading with what set it and since when, the biggest risk, and the link to the full dashboard), and the full card as a PNG image. as_of (YYYY-MM-DD, optional) returns the dashboard as Jawz would have read it that day, using only data published by then, with today's rules applied: rules_note says so and, when Jawz's daily snapshot for that date is on file, what Jawz published that day. as_of reaches back 365 days and excludes today, which is the default; a future date or one before the window returns an error stating the supported range. In a dated read every row carries point_in_time. The Investment card's Data centres and Chips rows are not point in time (the Census tables behind them keep only their revised figures, so a dated read shows those months as revised since) and point_in_time_note names them; every other row is as published by that date. The detail behind the cards is in get_macro_regime, get_financial_conditions, get_growth_indicators and get_inflation_indicators; get_indicator_history gives any card or row over time. Read-only, no account needed. Information, not investment advice.

  Parameters:

  • as_of — string — Optional past date, YYYY-MM-DD, within the last 365 days: the dashboard as Jawz would have read it that day, from data published by then. Omit (or pass today) for today's dashboard.

  Hints: read-only yes · destructive no · idempotent not declared · open-world no

• get_world_brief read-only

  Read the Jawz World Brief — the weekly Mako-bylined market read published at jawz.ai/brief. Returns the latest edition by default, or a specific one by slug (YYYY-MM-DD). Every claim in a brief traces to a Jawz tool read from its week. Setting list_only:true returns the available editions without their content.

  Parameters:

  • slug — string — Edition to fetch, YYYY-MM-DD. Omit for the latest.
  • list_only — boolean — Return only the list of available editions (slug, title, regimeLine, takeaway).

  Hints: read-only yes · destructive no · idempotent not declared · open-world no

• get_prices read-only

  Live prices for one or more holdings, and the answer shows its work. Resolves raw tickers (stocks, ETFs, crypto, international listings) and returns price, value, and day change per symbol, plus per holding: what it resolved to (asset: type, exchange, ISIN, provider id; asset.isin is filled when you pass an isin hint or the provider profile carries one, and is null otherwise — it is not looked up), how surely (resolution: matched_by isin|qualified_symbol|bare_symbol, candidates, mismatch against the caller's hints), the native quote with its own time (quote), the FX rate actually applied, dated and sourced (fx), and the position value native and in the base currency. Optional per-holding hints — isin (equities/ETFs/funds), exchange, asset_class — narrow resolution; a hint that contradicts the result is reported, not silently overridden. A bare ticker is resolved listing first: when the same symbol is both a US listing and a crypto token that is not a known major coin (HOOD, MSTR, and tokenized-stock proxies such as ALAB, LITE, NBIS), it prices the listing, never the token; the token is priced only when no such listing exists. A known major coin that shares its symbol with a well-known US listing (STX is Stacks and Seagate) stays the coin, with a line warning that names the listing and how to get it (qualify it, or asset_class 'equity'). A bare ticker shared by differently named funds on other venues (DRAM) is priced with a line warning naming the alternatives and their qualified symbols, counted in resolution.candidates. Qualifying as NASDAQ:X / X.US picks the US listing explicitly. Line warnings are in each holding's warning; a summary is in the top-level warnings. base_currency defaults to USD; 'native' returns unconverted quotes.

  Parameters:

  • holdings (required) — array — Portfolio holdings to analyze
  • base_currency — ['string', 'null'] — Currency for price, value and totalValue. Omit for USD (default). 'native' or null: no conversion — each line is stated in its quote currency and fx is null.

  Hints: read-only yes · destructive no · idempotent not declared · open-world yes

• get_macro_regime read-only

  The current macro regime (GREEN / YELLOW / RED) and business cycle (SPRING / SUMMER / FALL / WINTER) with the evidence behind them: the detail behind the Regime and Business cycle cards of get_dashboard. The regime combines consumer mood (U Michigan sentiment) with money (the global-liquidity condition), and the money condition changes only after two consecutive weekly reads agree: data.details.liquidity_condition shows this week's raw read, the held class, how many agreeing reads are still needed, and the margin to the line. The business cycle crosses growth with prices. Growth is a vote of factory output (IPMAN trend), jobs (initial claims) and investment (BEA real investment in information-processing equipment and software), two of three deciding, and the season changes only when two prints in a row agree: data.cycle carries the vote, the dated prints and held_since. The response opens with the Jawz Dashboard card: the six-reading card as markdown at the top of presentation.display_markdown (above a horizontal rule), the same object at data.read, and the card as a PNG image content block after the text. Below the card, display_markdown carries the regime write-up and its tables (regime inputs, the growth vote, the growth and inflation composites, asset-class tilts); 'tables' is their structured form and 'data' holds the raw values. A 'provenance' block lists each input with its as-of date and age in days. get_indicator_history gives the regime, the cycle or any of their rows over time.

  Hints: read-only yes · destructive no · idempotent not declared · open-world no

• get_regime_history read-only

  Macro regime read-history — a timeseries of how the regime and its drivers have moved over recent weeks. Returns one row per sample date (regime color, business-cycle quadrant, growth + inflation class, consumer sentiment, global liquidity, dominant risk, confidence) plus the transitions between them (e.g. SUMMER → FALL). Provides the observed history behind a 'what changed since …' question, rather than a comparison of two separate point reads. Rule changes that moved the published colour or season without the market moving are listed in data.events as method events (kind 'method'), each dated the first day the new rule was live, with what changed and why: for example the money condition waiting for two consecutive weekly reads (2026-09-25) and the growth vote with two prints to change the season (2026-09-30). Observed rows keep what Jawz published that day under the rules then in force. Reading note, enforced by the artifact: the Liquidity column steps for two reasons that are NOT the market, and both are listed as first-class events in summary.liquidity_events — a g4↔g3 coverage flip (the whole PBoC component entering or leaving the sum: not a liquidity move at all), and a China-leg publication step (PBoC publishes monthly ~15d in arrears, so the day a new statement is recorded the column absorbs a whole month of change: real change, wrong date). get_liquidity_history carries a constant-basis G3 series that is immune to both. The response includes a 'presentation' object whose 'display_markdown' is a pre-formatted table; 'data.rows' and 'summary' carry the same values structured. Each row is provenance-tagged: 'observed' (captured live that day — true vintage) or 'reconstructed' (computed point-in-time from vintage data — each series as it was published on that date, so later revisions are excluded; depth-limited). Defaults to the last 12 weeks, weekly. get_indicator_history covers any card or row of get_dashboard over a year, at its own print dates.

  Parameters:

  • interval — string — Sample cadence (default weekly)
  • lookback_weeks — number — How many weeks of history to cover (default 12, clamped 1–52)

  Hints: read-only yes · destructive no · idempotent not declared · open-world no

• get_liquidity_history read-only

  Global-liquidity read-history — a timeseries of central-bank liquidity over recent weeks. Returns one row per sample date with the full decomposition (Fed / ECB / BoJ / PBoC in USD trillions), the active basis (g4/g3/us_fallback), the supportive/neutral/draining classification of that week's read, and US net liquidity (us_net_t: Fed balance sheet minus the Treasury General Account minus reverse repo, USD trillions) with its own classification (us_net_class), plus coverage (basis) changes, China-leg publication steps, and classification changes as events. The held money condition, which changes only after two consecutive weekly reads agree, is in get_macro_regime and the Money card of get_dashboard. History companion to get_financial_conditions (Chapter 1 Mode 1.5 Global Liquidity Read). Reading note, enforced by the artifact: when the basis flips g4↔g3 (China PBoC publication freshness), the headline total moves by the ~$7T China component, which is a coverage change rather than a liquidity move. The window trend is therefore computed on constant G3 basis, and every basis flip is listed with an explicit note. A China-leg publication step is a DIFFERENT event and is reported separately: PBoC publishes monthly ~15 days in arrears, so on the day a new statement is recorded the PBoC leg absorbs a whole month of balance-sheet change (plus FX) in one step — real change, but not change that happened on that date, and not the same thing as a coverage flip. The response includes a 'presentation' object whose 'display_markdown' is a pre-formatted table; 'data.rows' and 'summary' carry the same values structured. Rows are provenance-tagged 'observed' (live that day, true vintage) or 'reconstructed' (point-in-time from vintage data — each series as it was published on that date, ~4 months deep). Defaults to the last 12 weeks, weekly. get_indicator_history covers any card or row of get_dashboard over a year, at its own print dates.

  Parameters:

  • interval — string — Sample cadence (default weekly)
  • lookback_weeks — number — How many weeks of history to cover (default 12, clamped 1–52)

  Hints: read-only yes · destructive no · idempotent not declared · open-world no

• get_conditions_history read-only

  Financial-conditions read-history — a timeseries of how conditions have moved over recent weeks. Returns one row per sample date with the composite (loose/neutral/tight), direction (easing/stable/tightening), and every pillar's value + class: 10Y real yield, HY/IG credit spreads, DXY, VIX, global-liquidity class. The rate legs behind the real yield are exposed numerically too (dgs10_pct, t10yie_pct, t5yie_pct, t10y2y_pct), so a real-yield move can be read as nominal-led or breakeven-led rather than only as a fused number. Completes the Chapter 1 history trio with get_regime_history (the judgment) and get_liquidity_history (the flow) — this is the price of risk. Surfaces trajectory that a point read hides (e.g. HY spreads widening for six straight weeks while VIX stays calm), and grounds 'conditions are tightening' statements in observed pillar changes. Window deltas per pillar are in 'summary.deltas'; composite/direction/pillar classification changes are listed as events. The response includes a 'presentation' object whose 'display_markdown' is a pre-formatted table; 'data.rows' and 'summary' carry the same values structured. Rows are provenance-tagged 'observed' (live that day, true vintage) or 'reconstructed' (point-in-time from vintage data — each series as it was published on that date, ~4 months deep). Defaults to the last 12 weeks, weekly. get_indicator_history covers any card or row of get_dashboard over a year, at its own print dates.

  Parameters:

  • interval — string — Sample cadence (default weekly)
  • lookback_weeks — number — How many weeks of history to cover (default 12, clamped 1–52)

  Hints: read-only yes · destructive no · idempotent not declared · open-world no

• get_indicator_history read-only

  Any card of the Jawz Dashboard (get_dashboard), or one of its rows, over time, each at its own print dates: the weekly reads for the weekly cards, the release dates of the series behind a row for the rest. With no arguments it returns the catalogue: every card and indicator key with its kind, cadence, unit, source and the earliest date it reaches. card alone returns that card's badge (indicator defaults to 'badge'); card plus indicator returns that row. from and to (YYYY-MM-DD) set the window: to defaults to today and from to one year before to; the history reaches back 365 days (store of value starts 84 days later), and a clamped window is echoed as requested_from and effective_from. Every point carries Jawz's word for that date and point_in_time. What Jawz computes (the six badges, the four-bank money total, the dollar effect, US money, the growth vote) also carries the value and its display at every point. Published series that Jawz passes through unchanged (mood, jobs, BEA investment, prices, rates, credit, the dollar, gold, bitcoin, data centres, chips, builders' capex, power) carry Jawz's word per print plus the source (provider, series, link and, for FRED series, the ALFRED vintage link), because the full numbers stay with the publisher. The answer adds the rule in one sentence, transitions (every change of word), runs, and the current word with the date it has held since (for a badge, the card's own held_since). History applies today's rules to what was published by each date (rules: 'today'): method_events lists rule changes in the window that moved the card, and published_differs lists, for the regime and business-cycle badges, the dates where what Jawz published that day differs. The four-bank money rows add the coverage basis, a constant three-bank total and coverage and PBoC-statement events. An unknown card or indicator, a malformed or future date, or a date before the window returns an error naming the valid keys or range. A cold window that runs out of compute time returns the points ready with partial: true and pending_from; the same call again continues from the points already kept. Read-only, no account needed. Information, not investment advice.

  Parameters:

  • to — string — Window end, YYYY-MM-DD. Defaults to today (UTC).
  • card — string — Card key: regime, money, business_cycle, financial_conditions, store_of_value or investment. Omit card and indicator for the catalogue.
  • from — string — Window start, YYYY-MM-DD. Defaults to one year before `to`; clamped to the 365-day window.
  • indicator — string — Indicator key within the card (the catalogue lists them), or 'badge'. Defaults to 'badge' when a card is given.

  Hints: read-only yes · destructive no · idempotent not declared · open-world no

• get_data_health read-only

  Freshness status for every data source Jawz reads: per-source release date, latest observation date, expected cadence, age in days, and whether any source is stale or unavailable. Also reports whether the ingestion pipeline is keeping up, and carries plain-language notices when a scheduled update has not arrived. Applies when the reliability of a figure depends on how current its underlying data is.

  Hints: read-only yes · destructive no · idempotent not declared · open-world no

• get_financial_conditions read-only

  Full financial conditions read, the detail behind the Money and Financial conditions cards of get_dashboard (and the regime's Money row): global liquidity (Fed + ECB + BoJ central-bank balance sheets, FX-converted to USD; PBoC included when a fresh Mako-curated publication is available — basis 'g4' vs 'g3'), the 10-year real yield with its nominal and breakeven legs, HY/IG credit spreads, the dollar index (DXY), VIX. Inside global liquidity: us_net_liquidity (Fed balance sheet minus the Treasury General Account minus overnight reverse repo, with 4- and 12-week changes, its own classification and dated inputs; a supporting read, not part of the regime light) and change_decomposition (for the 4- and 12-week windows, how much of the change in dollars came from the balance sheets themselves and how much from exchange rates, which dominated, and the method). Returns composite + direction + per-pillar classifications + drivers. The `scope` and `scope_note` fields tell you whether the read is G4 or G3, and the China sub-component carries Mako-curated provenance (as_of_month, source_url, note). Pass summary_only:true for the lightweight one-line read used in regime composition.

  Parameters:

  • summary_only — boolean — If true, return composite + direction + one-line driver only

  Hints: read-only yes · destructive no · idempotent not declared · open-world no

• get_growth_indicators read-only

  Growth pillar indicators, the detail behind the Business cycle card of get_dashboard (its Growth, Jobs and Investment rows) and the regime's Mood row: industrial production (IPMAN, the ISM proxy), consumer sentiment (U Michigan), the 2s10s yield curve, initial claims, plus the Atlanta Fed GDPNow nowcast of the quarter in progress. growth_vote_inputs holds the three inputs of the business cycle's growth vote, each with its word and dated observation: output (IPMAN three-month change), jobs (initial claims four-week average against twelve weeks earlier) and investment (BEA real private fixed investment in information-processing equipment and software, A679RX1Q020SBEA, year on year, with its quarter), plus the vote's rule; the vote itself is applied in get_macro_regime. GDPNow carries its own reference_quarter, the publication date of the estimate, and the previous estimate of the SAME quarter, so its direction of travel is readable; it is reported, not scored, so it does not move the classification. When no estimate is on file the block reports status: unavailable and says so in staleness_flags — not published, which is not the same as zero. Returns classification (green/yellow/red), the growth pillar's own score, separate from the vote; read_complete is false when a series failed to read.

  Hints: read-only yes · destructive no · idempotent not declared · open-world no

• get_inflation_indicators read-only

  Inflation pillar indicators, the detail behind the Prices row of the Business cycle card of get_dashboard: headline + core CPI, headline + core PCE (YoY and MoM), 5Y/10Y breakevens, wage growth. Returns classification (supportive/neutral/headwind).

  Hints: read-only yes · destructive no · idempotent not declared · open-world no

• get_event_calendar read-only

  Scheduled US macro events (FOMC + CPI + NFP + PCE) with dates and consensus. Consensus values are Mako-curated and arrive with provenance (source, source_url, published_at, age_days). An event with no curated consensus carries consensus: null, which means the value is not published rather than zero. Mode 1.2 covers the qualitative-scenario path for events whose consensus is not published. consensus_metrics_expected lists the metric keys a complete consensus would carry for that event type. Market-implied pricing (what_is_priced / CME FedWatch) is a separate quantity and remains deferred. COVERAGE: only fomc, cpi, nfp and pce have dates in this calendar. ecb, boj and boe are accepted for consistency with the consensus vocabulary but carry NO dates, so filtering to them returns an empty list and says so in staleness_flags — that is 'not tracked here', never 'none is scheduled'. Coverage also runs to a fixed last date; a window past it is flagged rather than silently short.

  Parameters:

  • event_types — array — Filter by event types (default: fomc, cpi, nfp, pce). Only those four carry dates; ecb/boj/boe return empty with a staleness flag.
  • lookahead_days — number — How many days forward to look (default 14)

  Hints: read-only yes · destructive no · idempotent not declared · open-world no

• get_weekly_data_releases read-only

  Past week's data releases with H.4.1 net liquidity update. Anchored to Friday — most current when run Friday morning or later. Each release classifies actual vs consensus where BOTH are on file (both Mako-curated): `surprise` is hot / modestly hot / in-line / modestly cool / cool on the headline metric, `vs_consensus` is the same comparison stated neutrally as above/below/in-line, and `metrics` breaks it down per metric. hot/cool is DIRECTIONAL versus consensus, not a verdict — a hot CPI and a hot payrolls print mean opposite things for the same book. `surprise: "n/a"` means the comparison could not be made; `implication` names which half is missing. consensus_provenance.age_days is measured at week_ending, not at call time, so re-asking for an earlier week returns the same age it did the first time.

  Parameters:

  • week_ending — string — Week ending date (YYYY-MM-DD); defaults to last Friday

  Hints: read-only yes · destructive no · idempotent not declared · open-world no

• get_etf_profile read-only

  ETF look-through for Chapter 2 Mode 2.2 (Concentration Check). Given ETF ticker(s), returns each fund's top holdings with weights so the AI can overlay them with the user's direct positions and surface hidden single-name concentration. Each profile carries source='live' (Vanguard API, fresh) or source='catalog' (dated snapshot — see as_of). Holdings are TOP-N only, so any true-exposure figure computed from this is a floor, not exact. Bond and commodity ETFs return no holdings by design. Tickers not in the catalog come back in unknown_tickers; their contents are not known to Jawz and no holdings are returned for them.

  Parameters:

  • tickers (required) — array — ETF ticker symbols to look through (e.g. ['QQQ','VGT','SPY'])

  Hints: read-only yes · destructive no · idempotent not declared · open-world yes

• score_position_drift read-only

  Score how a single position's regime fit has drifted since entry. Returns regime_at_entry, regime_now, fit_score_at_entry, fit_score_now, drift_score (now - entry), drift_label (improved/stable/deteriorated/review_now), explanation, and review_questions. No buy/sell recommendation — output is observational. Supports Chapter 4 Mode 4.2 (Position Retrospective) for single-name regime-fit review and Mode 4.3 (Thesis Status Sweep) for per-position drift across the book.

  Parameters:

  • symbol (required) — string — Ticker or instrument (e.g. AAPL, BTC, TLT)
  • asset_class (required) — string — Asset class for fit-score lookup, drawn from the holding's actual exposure (Chapter 2 Mode 2.1) rather than from the current regime. A hybrid holding is scored per sleeve and weight-blended. Drives the regime-fit calculation.
  • entry_date (required) — string — ISO date (YYYY-MM-DD) when the position was opened
  • thesis — string — Original thesis text. Echoed back word-for-word in review questions.
  • conviction — number — User-supplied conviction 1–10. Drives conviction_gap if a meaningful gap exists vs the regime fit score.
  • symbol_type — string — Price-API hint. Inferred from asset_class if omitted.

  Hints: read-only yes · destructive no · idempotent not declared · open-world no

• get_drift_alerts read-only

  Portfolio-wide drift scan. Iterates Decision Records, calls score_position_drift on each, returns flagged positions with severity + why_flagged + review_questions. Flag types: drift_severe (review_now), drift_meaningful (deteriorated), fit_low_structural (score ≤2 but unchanged since entry — low by design, not decay), fit_low_decayed (score ≤2 AND below entry fit — deteriorated after entry), conviction_gap (|gap| ≥3), thesis_undocumented (entry_date or thesis missing). Backs Chapter 4 Mode 4.3 (Thesis Status Sweep) as the portfolio-wide drift scan; flagged names feed a Mode 4.2 (Position Retrospective).

  Parameters:

  • portfolio (required) — array — Array of Decision Records — one per position the user wants scanned

  Hints: read-only yes · destructive no · idempotent not declared · open-world no

• list_loops read-only

  List all available Jawz loops. Returns loop metadata, access status, and which loop is currently active for this user. Every published loop is free and open. Deprecated loops you are entitled to are included with deprecated: true.

  Hints: read-only yes · destructive no · idempotent not declared · open-world no

• get_loop read-only

  Get metadata and chapter manifest for a loop by slug. Returns loop info and the list of chapter nodes with titles — no chapter content. All active loops are navigable by anyone.

  Parameters:

  • loop_slug (required) — string — Loop slug (e.g. 'jawz')

  Hints: read-only yes · destructive no · idempotent not declared · open-world no

• get_chapter read-only

  Fetch the full content of a chapter by loop slug and node. Loops not marked open require entitlement; every published loop is currently open and free. Does not log an invocation; run_chapter is the variant that runs the framework and logs.

  Parameters:

  • loop_slug (required) — string — Loop slug
  • node (required) — string — Chapter node

  Hints: read-only yes · destructive no · idempotent not declared · open-world no

• run_chapter

  Returns a chapter's published framework text: its modes, the questions each asks, and the output format it describes. The framework is reference material for the assistant to apply as it judges appropriate. Repeat calls return the same chapter content — the text is not regenerated per call. Each call appends one usage-counter row and returns a new invocation_id used for feedback.

  Parameters:

  • loop_slug (required) — string — Loop slug
  • node (required) — string — Chapter node
  • context — string — Optional user context (portfolio details, ticker, question)

  Hints: read-only no · destructive no · idempotent not declared · open-world no

• run_mode

  Returns the published framework text for one named mode within a chapter, rather than the full chapter. Repeat calls return the same mode content — the text is not regenerated per call. Each call appends one usage-counter row and returns a new invocation_id used for feedback.

  Parameters:

  • loop_slug (required) — string — Loop slug
  • node (required) — string — Chapter node
  • mode_slug (required) — string — Mode slug (e.g. 'equity-screen', 'regime-overlay')
  • context — string — Optional user context

  Hints: read-only no · destructive no · idempotent not declared · open-world no

• get_china_liquidity read-only

  Read the latest Mako-curated PBoC balance-sheet publication (CNY trillions + source URL + as_of_month + note). Public read — suitable for surfacing directly when explaining the China leg of global liquidity. The same value is also embedded in get_financial_conditions.pillars.global_liquidity.components.china and is the PBoC leg of the Money card of get_dashboard; this tool exists for direct/auditable access.

  Hints: read-only yes · destructive no · idempotent not declared · open-world no

Publisher website

Measured by MCPYP probing. Untested fields are omitted, never estimated. Listing is always free; MCPYP never proxies calls or stores user tokens.