Open Economics API

A free, read-only semantic routing layer for authoritative Brazilian economic data.

• Semantic API: https://open-economics-data.knbf982hkn.chatgpt.site/api/v2
• Stable series API: https://open-economics-data.knbf982hkn.chatgpt.site/api/v1
• MCP: https://open-economics-data.knbf982hkn.chatgpt.site/api/mcp
• Ask the data: https://open-economics-data.knbf982hkn.chatgpt.site/en/ask
• MCP setup: https://open-economics-data.knbf982hkn.chatgpt.site/en/mcp
• Agent installer: llms-install.md
• Documentation: https://open-economics-data.knbf982hkn.chatgpt.site/en/docs
• Runnable examples: https://open-economics-data.knbf982hkn.chatgpt.site/en/guides
• Source and issue tracker: https://github.com/felipegambettadesouza6-jpg/open-economics
• Reliability policy: RELIABILITY.md
• 2.0 release evidence: benchmarks/RELEASE-READINESS.md
• Contributing: CONTRIBUTING.md
• Security: SECURITY.md

Run a real economic question → · Connect the MCP server →

Open Economics 2.0 starts with a real economic-information need, resolves its meaning independently of current coverage, and then routes it to official data. Its synchronized catalog currently exposes 12,875 BCB SGS series and IBGE aggregates, plus direct SICONFI fiscal-report, MDIC Comex Stat, and ANP fuel-price access and versioned EPE electricity-consumption, MTE Novo Caged, CVM investment-fund, Tesouro RTN fiscal, and RMD Federal Public Debt data (12,886 official datasets in total). The 32 convenient v1 series IDs remain compatible.

Discovery, REST, the existing product, and MCP share one semantic core. A resolved concept is kept separate from availability, so an unsupported need is reported explicitly instead of being silently mapped to a nearby series. Units, dimensions, reference periods, source identifiers, raw values, methodology links, and retrieval provenance travel with the data.

Start with the economic need

curl --fail --silent \
  "https://open-economics-data.knbf982hkn.chatgpt.site/api/v2/search?q=desemprego%20desde%202015"

curl --fail --silent \
  "https://open-economics-data.knbf982hkn.chatgpt.site/api/v2/datasets/ibge-aggregates%3A6381/schema"

For BCB, /api/v2/datasets/bcb-sgs:{code}/observations provides direct series access with authoritative frequency, unit, source, coverage, formula, and warning metadata. For IBGE, inspect /schema, then send explicit variable, periods, locality, and classification selections to /observations. SICONFI DCA, RREO, and RGF use the same dataset routes while retaining entity, reporting period, annex, account, column, and raw value. Official multidimensional structure is preserved instead of flattened away. Comex Stat preserves trade-flow dimensions and metrics. ANP fuel-price queries return period/geography/product aggregates calculated from official station observations, with source counts and the transformation disclosed while station identity and address fields are excluded. EPE and MTE serve compact, versioned snapshots of official workbooks: electricity retains geography/class/market, while adjusted Novo Caged stock and flows retain their separate national, region/state, or economic-activity breakdowns. CVM daily fund reports retain fund/class identity and quota values; classification aggregates sum only additive measures and identify incomplete filing dates. RTN retains its monthly account hierarchy and above-the-line cash/effective-payment conventions. RMD debt statistics keep composition, holder, maturity, and cost tables separate, with their official units, definitions, and publication vintage.

When to use Open Economics

If you already know the exact official identifier and source contract, calling the publisher directly remains the shortest path. Open Economics is useful when the need starts in human language, spans publisher conventions, requires explicit dimensions, or must retain one consistent provenance and error model.

One contract across official sources

The same observation envelope can retrieve BCB IBC-Br (SGS 24363) and IBGE real GDP growth (SIDRA 5932/6561), without maintaining two date and response parsers:

curl --fail --silent "https://open-economics-data.knbf982hkn.chatgpt.site/api/v1/indicators/br-ibc-br/observations?start=2024-01-01"
curl --fail --silent "https://open-economics-data.knbf982hkn.chatgpt.site/api/v1/indicators/br-gdp-real-yoy/observations?start=2024-01-01"

The runnable multi-source Python example uses both series and prints their official provenance and freshness state.

Start with the catalog

curl --fail --silent "https://open-economics-data.knbf982hkn.chatgpt.site/api/v1"
curl --fail --silent "https://open-economics-data.knbf982hkn.chatgpt.site/api/v1/indicators?q=ipca&source=ibge"
curl --fail --silent "https://open-economics-data.knbf982hkn.chatgpt.site/api/v1/indicators/br-ipca-monthly"

GET /api/v1/indicators is the discovery endpoint. Its meta.available_filters field lists every canonical category, frequency, and source value. It accepts:

Invalid filters are rejected with a structured problem response; they never silently become an empty result set.

Retrieve a series

curl --fail --silent \
  "https://open-economics-data.knbf982hkn.chatgpt.site/api/v1/indicators/br-ipca-monthly/observations?start=2024-01-01&end=2024-12-31"

curl --fail --silent \
  "https://open-economics-data.knbf982hkn.chatgpt.site/api/v1/indicators/br-selic-target/observations?start=2025-01-01&order=desc&limit=12"

curl --fail --silent \
  "https://open-economics-data.knbf982hkn.chatgpt.site/api/v1/indicators/br-selic-target/latest"

Observation requests accept start, end, order=asc|desc, limit=1..5000, and format=json|csv. Dates use YYYY-MM-DD and end cannot be in the future. Daily BCB requests are limited to ten years because SGS applies the same upstream limit.

All JSON series responses use the same envelope:

{
  "data": [
    {
      "date": "2024-01-01",
      "period": "2024-01",
      "source_date": "202401",
      "value": 0.42,
      "raw_value": "0.42",
      "status": "observed"
    }
  ],
  "meta": {
    "indicator": { "id": "br-ipca-monthly", "unit_symbol": "%" },
    "provenance": { "upstream_url": "…", "retrieved_at": "…" },
    "returned": 1,
    "available": 1,
    "truncated": false
  }
}

date is the normalized start date for the reference period; period is the frequency-aware identifier (YYYY-MM-DD, YYYY-MM, YYYY-QN, or YYYY). source_date and raw_value are retained exactly from the official publisher. Read meta.indicator.date_semantics before interpreting stock, flow, or moving-quarter series.

Use format=csv for a flat download. CSV rows repeat indicator_id, source_id, source_url, and upstream_url, so exported values retain their provenance outside the JSON envelope.

API surface

All endpoints support CORS and GET, HEAD, and OPTIONS. Successful and error responses include X-Request-Id; browser clients can also read cache, timing, and stale-response headers.

Errors and freshness

Errors use application/problem+json with a stable code, HTTP status, human-readable title, explanatory detail, and request_id. Common cases include INVALID_DATE, INVALID_CATEGORY, INDICATOR_NOT_FOUND, UPSTREAM_CONNECTION_ERROR, and UPSTREAM_TIMEOUT.

The API caches successfully normalized source responses in D1 when configured. If a refresh fails and a previous matching snapshot exists, it is returned with meta.stale: true, meta.cache: "stale", and HTTP Warning: 110. A cache read or write failure is treated as a cache bypass, never as a data failure.

The service is currently best-effort and has no uptime SLA. See RELIABILITY.md for the explicit availability, freshness, change-management, and incident-reporting policy.

Sources and correctness