Isoba — Dollar & Crisis Index

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

US dollar index, crisis meter, recession and market-top risk from public-domain data since 1990.

Add to your agent

Copy this into your MCP client config:

{
  "mcpServers": {
    "Isoba — Dollar & Crisis Index": {
      "url": "https://isoba.lokio.org/mcp",
      "transport": "streamable-http"
    }
  }
}

Measured tools (8)

• get_current_readings read-only

  Today's readings with the data date: Dollar System Index (where the dollar system stands and what it is made of), Crisis meter (stress already present in credit and funding, four sensors with week-ago and month-ago values), market-top risk and US recession risk (how often an S&P 500 bear market or a US recession followed the same zone since 1990, with the number of independent episodes behind each number), what changed this week, and a link to the public daily record. Say the data date and credit "Isoba (LOKIO)". Describe readings as where things stand now and as history ("in similar periods since 1990, X happened in N of M cases"), never as a forecast, a signal, or advice to buy or sell. If someone asks what to do with their own money, give the readings and say that this is not investment advice.

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

• get_readings_on_date read-only

  What Isoba showed on any working day since 1990-01-01: the four instruments with zones, every sensor, the eight filters (since 2006), the yield curve and the four credit inputs — computed point-in-time from data published by that day. Use for "what did it show before 2008 / on the day of COVID?".

  Parameters:

  • date (required) — string — YYYY-MM-DD; a weekend or holiday gives the previous working day

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

• get_changes_since read-only

  What moved between two dates and why: each instrument (from → to, zone change), the sensors that moved most and any that switched on or off, and the filters. Use when the user asked before ("since last week / since I last checked") or asks why a reading changed.

  Parameters:

  • since (required) — string — Start date YYYY-MM-DD
  • to — string — End date YYYY-MM-DD (default: latest)

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

• find_similar_days read-only

  Days since 1990 when the 14 sensors of the crisis meter, market-top risk and recession risk stood closest to a given day (default: today), at least half a year apart and at least a year before the given day, and what followed within 12 months: the crisis meter peak, a US recession (NBER), an S&P 500 bear market, the dollar index. History, not a forecast — always show the spread of outcomes.

  Parameters:

  • date — string — Day to compare, YYYY-MM-DD (default: latest)
  • count — integer — How many days (default 5)

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

• what_to_watch read-only

  How many points each instrument is from the next zone boundary up or down, which sensors are closest to switching on or off (threshold 80), and the freshest and stalest inputs. Use for "what would change the picture?".

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

• get_expectations read-only

  What markets price now (Fed rate path from the 2-year Treasury yield, 5-year breakeven inflation, yields) and what happened over the next 12 months after the 10 days since 1990 most similar to today (same 14 sensors as find_similar_days, at least half a year apart), with the number of distinct cycles behind them: US dollar, US stocks (market value of companies, Fed Z.1), 10-year Treasury return, Fed rate, core inflation, unemployment, WTI oil, yield curve, corporate bond spread, cash vs inflation, M2 money growth, and the crypto backdrop (dollar and M2 — there is no open crypto price history since 1990). Report counts ("rose in 9 of 10"), the median and the range, and how many cycles stand behind them. History and market prices, not a forecast. Say the data date and credit "Isoba (LOKIO)". Describe readings as where things stand now and as history ("in similar periods since 1990, X happened in N of M cases"), never as a forecast, a signal, or advice to buy or sell. If someone asks what to do with their own money, give the readings and say that this is not investment advice.

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

• get_monthly_history read-only

  Monthly averages of one or several instruments between two months (DSI since 2006, the others since 1990), with the zone of each month. Use for trends and to see what the instruments showed around past events.

  Parameters:

  • to — string — Last month, YYYY-MM (default: latest)
  • from — string — First month, YYYY-MM (default: 24 months ago)
  • instrument — string — dsi — Dollar System Index; crisis — Crisis meter; top_risk — market-top risk; recession_risk — US recession risk
  • instruments — array — Several instruments at once (instead of instrument)

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

• get_track_record read-only

  How Isoba behaved when run point-in-time over every working day since 1990 (only data published by each day, checked for look-ahead). Default section "summary" is short; ask for a section (crisis_events, recessions, market_tops, false_alarms, by_year, methods, all) or one event (gulf, ltcm, dotcom, sept11, worldcom, bnp, lehman, euro, downgrade, china, oil2016, q4_2018, covid, bear2022, svb, tariffs — or its year). Call it before claiming reliability.

  Parameters:

  • event — string — One crisis by id, name or year, e.g. lehman, "COVID", 2008
  • section — string — Part of the record (default summary)
  • to_year — integer
  • from_year — integer

  Hints: read-only yes · destructive not declared · idempotent yes · 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.