Banks

Every US bank here is an FDIC-insured institution identified by its certificate number — the cert. It is not a ticker, not an RSSD id, and not a CIK: a bank holding company that files with the SEC and the insured bank underneath it are different entities with different identifiers. So the first call is almost always a resolution call, and the second is the read.

Financials arrive as report quarters, newest first, with values keyed by a curated metric_key. Each value carries its own exact decimal, unit, currency, and as_of, and data.metric_definitions explains the vocabulary you asked for: title, units, currency, basis, category, and a one-line description. An agent can therefore interpret a metric it has never seen before without a lookup table baked into its prompt.

Resolve a bank name to a certificate#

banks_search_institutions(query="Frost Bank", limit=5)

Ranked matches come back with both identifiers, location, active status, regulator and charter metadata, lifecycle dates, a score, and the reason the row matched:

{
  "data": {
    "query": "Frost Bank",
    "matches": [
      {
        "cert": 99999,
        "rssd_id": 999999,
        "name": "<institution legal name>",
        "city": "<city>",
        "state": "TX",
        "active": true,
        "primary_regulator": "FDIC",
        "charter_class": "SM",
        "established_date": "1899-01-01",
        "closed_date": null,
        "score": 100,
        "match_reason": "exact_name",
        "as_of": "2026-08-19T00:00:00Z"
      }
    ]
  },
  "meta": { "source": "fdic", "as_of": "2026-08-19T00:00:00Z" },
  "pagination": null
}

Values are illustrative; field names are real. pagination is null because a relevance list is bounded by construction rather than paged. Digits in the query are treated as an exact identifier lookup, so passing a cert or an RSSD id straight through works. A name that matches nothing comes back as unknown_entity, sometimes with suggestions attached.

Read a bank's quarterly financials#

Take the cert from the search and ask for the metrics you actually need:

banks_get_financials(cert=628, metrics=["total_deposits", "re_construction_land"], limit=4)

{
  "data": {
    "cert": 628,
    "name": "<institution legal name>",
    "rssd_id": 999999,
    "active": true,
    "city": "<city>",
    "state": "NY",
    "period_type": "quarter",
    "quarters": [
      {
        "period": "2026-06-30",
        "fiscal_quarter": "2026Q2",
        "values": {
          "total_deposits": {
            "value": "2000000.000",
            "unit": "USD thousands",
            "currency": "USD",
            "as_of": "2026-08-19T00:00:00Z"
          }
        }
      }
    ],
    "metric_definitions": {
      "total_deposits": {
        "title": "Total deposits",
        "units": "USD thousands",
        "currency": "USD",
        "basis": "period_end",
        "category": "size",
        "description": "..."
      }
    }
  },
  "meta": { "source": "fdic", "as_of": "2026-08-19T00:00:00Z" },
  "pagination": { "limit": 4, "has_more": true, "next_cursor": "eyJ2IjoxLCJrIjpbLi4uXX0" }
}

Omit metrics entirely to get the whole curated vocabulary for those quarters — useful once, to see what exists, and wasteful on every call after that.

Walk a bank back through its history#

Quarters page like everything else: keep cert, metrics, start, and end constant, pass the previous response's cursor, and stop when has_more is false.

banks_get_financials(cert=628, metrics=["total_deposits"], limit=8, cursor="eyJ2IjoxLCJrIjpbLi4uXX0")

For a fixed analysis window, bound it instead of paging blindly:

banks_get_financials(cert=628, metrics=["total_deposits"], start="2024-01-01", end="2026-06-30")

start and end filter on the FDIC quarter-end report date, inclusive at both ends. A reversed window is a bad_parameter, as is a metric key that is not in the curated catalog.

• Monetary values are USD thousands, not dollars. Every currency metric in this category is reported the way the FDIC reports it: multiply by 1,000 before you say "dollars" in a sentence a human will read. The unit rides on each value, so there is no excuse for guessing.
• Some metrics are year-to-date flows, not quarterly ones. A definition whose basis is ytd_flow is cumulative from the start of that calendar year. Differencing consecutive quarters gives you the quarterly figure; treating the raw value as a quarter overstates it late in the year. Percent metrics carry a null currency.
• This is the insured bank, not the holding company. Consolidated group financials are a different filing at a different regulator and are not what these tools return.
• Coverage is the FDIC directory. An unrecognised cert is unknown_entity; a well-formed request for an institution we do hold, with nothing in the window, is a normal empty success.