MCP server reference

get_data_catalog

Lists every data source the account can read, plus the table structure that all sources share. Call it once at the start of a session.

Returns. The visible schemas with their organization and country, a list named schemas_without_data for schemas that hold no rows yet, a dataset index per schema, and the definitions of the series, data_points, dimensions and compound_series tables.

Example arguments

{
  "schemas": "bls,bea"
}

search_datasets

Keyword search over dataset titles and topics in every schema. The ranking is by keyword, not by meaning.

Returns. Ranked rows of schema, dataset_code, title and topic.

Example arguments

{
  "query": "consumer prices",
  "schemas": [
    "bls"
  ]
}

describe_dataset

Full metadata for one dataset. Call it after search_datasets has identified the dataset.

Returns. Topic, methodology, known breaks in the series, any notice of a changed base year, the dimensions that exist in the data, and example series.

Example arguments

{
  "schema": "bls",
  "dataset_code": "cu"
}

search_series

Finds series in one schema by words in the title. The match is a substring match, so short word stems work best ("rare" finds more than "rare earth").

Returns. Matching series with series_id, title, dataset_code, frequency, units, seasonal adjustment, begin_time and end_time, plus a note that says whether all terms or any term matched.

• When this tool finds nothing, query the series and dimensions tables with run_sql.

Example arguments

{
  "schema": "bls",
  "terms": [
    "unemployment",
    "rate"
  ]
}

get_series

Returns one series with its metadata. It accepts ordinary time series, tabular series and COMPOUND:: series.

Returns. Title, units, frequency, notes, source metadata and the observations. A coverage_note and a missing_periods list appear when the series has periods with no observation. Series built from company filings also carry row_sources, which names the filing behind each row.

Example arguments

{
  "schema": "bls",
  "series_id": "LNS14000000",
  "from_year": 2024,
  "transform": "yoy_diff"
}

run_sql

Runs one read-only SELECT statement against one schema. Use it for joins, pivots, aggregation, and catalog searches that the search tools cannot express.

Returns. row_count, columns and results. When the statement matches more than 50 rows, truncated is true and a note says how many rows exist. On nasa_fires the result carries page and has_more instead.

• The database role is read-only. INSERT, UPDATE, DELETE and DDL statements are rejected.
• A statement is cancelled after 30 seconds.
• For a year-over-year comparison, join on the calendar period with date_trunc. Do not use LAG(value, 12): one missing month makes every later comparison wrong.

Example arguments

{
  "schema": "bls",
  "sql": "SELECT date_trunc('year', time) AS year, avg(value) AS unemployment_rate FROM bls.data_points WHERE series_id = 'LNS14000000' AND time >= '2015-01-01' GROUP BY 1 ORDER BY 1"
}

get_market_data

Recent prices and company profiles from a market-data provider. Use it for quotes, currencies and commodities that the stored series do not yet cover.

Returns. Price rows with open, high, low, close and volume, or the requested quote or profile.

• Financial statements and reported company figures come from search_company_filings, not from this tool.

Example arguments

{
  "asset": "EUR/USD",
  "data_type": "price_history",
  "frequency": "weekly",
  "limit": 52
}

get_geo_data

Satellite-derived indicators for a country, one state or province, or a bounding box. FactIQ computes the result from the provider at the time of the call, so a call can take 30 seconds or more.

Returns. Rows for each interval or grid cell, with units, the source attribution to cite, and caveats for the dataset.

• State and province boundaries exist for IND, CHN, IDN, VNM, THA, MYS, PHL, PAK, BGD, LKA, MMR, KHM, NPL, KOR, JPN, TWN and USA.
• Most datasets allow at most 50 intervals per call. fires_viirs allows 200.

Example arguments

{
  "dataset": "fires_viirs",
  "region": "India/Punjab",
  "start_date": "2025-10-01",
  "end_date": "2025-11-30",
  "aggregation": "daily"
}

search_company_filings

Figures and statements from the reports one company filed: US SEC filers, and companies listed in Germany, the UK and France.

Returns. A tree grouped by filing or by measure. Each filing and each figure carries source_url and a source_link object that points to the original document.

• commentary and risk_changes cover US SEC filers only.
• When an exact ticker has no stored figure for a metrics or facts request, the tool answers from a market-data provider’s standardized statements. Those values carry no link to a filing.

Example arguments

{
  "company": "NVDA",
  "search_target": "facts",
  "concept": "revenue",
  "report_type": "annual"
}

search_earnings_transcripts

What management and analysts said on earnings calls, stored as separate claims with an exact quote each. It never returns a whole transcript.

Returns. Rows with the speaker, role, fiscal period, a normalized statement, the exact quote, and a source_link. Coverage rows give the number of calls and the first and last period for each company.

• Only verbatim_quote is a quotation. canonical_statement is a rewritten summary and must not be placed in quotation marks.
• An empty result does not prove that management said nothing on the subject. Check coverage first, then try other words.

Example arguments

{
  "search_target": "claims",
  "ticker": "MU",
  "query": "capital expenditure",
  "detail": true
}

search_media_appearances

Claims made on podcasts, in television interviews and at conferences, and in official documents such as Federal Reserve speeches, testimony, statements and minutes.

Returns. Rows with the speaker, a summary of what was said, topic labels, the channel and publication date, and a source_url that opens the video at the right time or the document at the right page.

• The text in canonical_paraphrase is a summary, not a quotation. Open source_url and check the wording before you quote it.
• A company value that matches nothing is listed under company_unmatched with up to five possible matches. It returns no rows and no error.

Example arguments

{
  "search_target": "claims",
  "company": "OpenAI",
  "query": "compute",
  "sort": "newest"
}

search_news

Headlines and short publisher summaries from the news feeds FactIQ collects. It holds recent news, not an archive, and it never returns the body of an article.

Returns. rows with published_at, source, title, summary, url, the tickers named in the text, and an analysis block with keywords, a geography and one sentence on why the story matters. meta gives the number matched and returned.

Example arguments

{
  "query": "copper",
  "topic": "markets",
  "limit": 10
}

get_style_guides

Returns FactIQ’s written guides for charts, reports, SQL and earnings-call work as Markdown text.

Returns. One Markdown document per requested guide, keyed by guide name.

Example arguments

{
  "guides": [
    "chart",
    "sql"
  ]
}

render_chart

Draws an interactive line, bar or area chart inside the conversation from numbers the assistant has already fetched. The server lists this tool only for ChatGPT and Codex, which can display the chart component.

Returns. The chart specification, which the client draws. The tool does not create a public page or a link.

Example arguments

{
  "title": "US unemployment rate held near 4.1% through mid-2026",
  "x_values": [
    "2026-06-01",
    "2026-07-01",
    "2026-08-01"
  ],
  "series": [
    {
      "label": "Unemployment rate",
      "values": [
        4.2,
        4.1,
        4.1
      ]
    }
  ],
  "units": "Percent"
}

send_feedback

Sends a report about wrong data or a failing tool to the FactIQ team. It is the only tool that does more than read.

Returns. A confirmation that the report was stored. No reply comes back through the connector.

Example arguments

{
  "category": "data_issue",
  "message": "bls LNS14000000: get_series returns no row for 2026-05-01, but the BLS website publishes a value for May 2026."
}