The full upstream README, mirrored here for reference. Install config, tool schemas, adoption signals, and an original overview live on the Open Economics listing page.
A free, read-only semantic routing layer for authoritative Brazilian economic data.
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.
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.
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.
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:
The runnable multi-source Python example uses both series and prints their official provenance and freshness state.
GET /api/v1/indicators is the discovery endpoint. Its
meta.available_filters field lists every canonical category, frequency,
and source value. It accepts:
| Parameter | Meaning |
|---|---|
q | Case-insensitive search across IDs, names, aliases, and official codes |
category | A canonical category ID such as inflation or interest-rates |
frequency | daily, monthly, quarterly, or annual |
source | bcb or ibge |
limit | 1–500, default 100 |
Invalid filters are rejected with a structured problem response; they never silently become an empty result set.
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:
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.
| Endpoint | Purpose |
|---|---|
GET /api/v1 | Machine-readable API discovery |
GET /api/v1/indicators | Search and filter the indicator catalog |
GET /api/v1/indicators/:id | Full indicator metadata, units, semantics, and links |
GET /api/v1/indicators/:id/observations | Normalized historical values |
GET /api/v1/indicators/:id/latest | Latest available observation |
GET /api/v1/sources | Publisher, attribution, and license metadata |
GET /api/v1/openapi.json | OpenAPI 3.1 description |
GET /api/v1/health | Router and catalog readiness (does not call publishers) |
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 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.
status values.There are 32 curated indicators across inflation, interest rates, currencies, activity, labor, credit, fiscal, external, and markets. Values are never fabricated, forward-filled, or silently sign-inverted. BCB NFSP fiscal series retain BCB's financing-requirement sign convention.
BCB catalog data is published under ODbL; preserve its attribution and share-alike obligations when distributing adapted databases. Attribute IBGE as IBGE/SIDRA and retain its source links and terms. Each indicator response has the authoritative URLs and license applicable to that series.
Node.js 22.13+ is required.
Examples are available in examples/python.py,
examples/multi_source_python.py, and
examples/javascript.mjs. The D1 migration in
drizzle/ is the deployment record for cache snapshots.
The API implementation is released under the MIT License. Upstream data remains governed by each publisher's own terms and licenses.