OpenArg — Argentina public data

What OpenArg — Argentina public data MCP server does

OpenArg — Argentina public data MCP server gives an MCP-compatible assistant access to more than 33,000 datasets published through 38 Argentine government portals. Coverage includes INDEC, the Central Bank of Argentina, national ministries, provincial and municipal sources, and Congress. The service returns dataset titles, portal information, official source links, and table data or metadata where available.

The server supports two usage patterns. In data mode, the assistant searches the catalog, examines table structure, retrieves selected rows, and performs calculations. In answers mode, consultar_datos_publicos answers a question in natural language, selects and cross-references relevant tables, and includes sources, links, data-quality or coverage warnings, and remaining answer quota.

How it works

The hosted MCP endpoint is https://mcp.openarg.org/mcp and uses Streamable HTTP. Clients send an Authorization: Bearer oarg_sk_... header with a key obtained from the OpenArg developer page. The service is read-only: its tools search and analyze published information but do not modify government datasets.

For reproducible table work, begin with buscar_datasets, then use describir_tabla to confirm column names, types, row count, date coverage, freshness, and a sample. Retrieve matching rows with obtener_datos, or use agregar_datos for sums, averages, counts, minimums, maximums, and grouped rankings. Aggregation runs in the database and avoids the 500-row retrieval limit.

Setup and configuration

Add the hosted endpoint to an MCP client with the API key in the authorization header. The README provides configurations for Claude Code, Cursor, Claude Desktop through mcp-remote, and VS Code. A custom agent can use the same endpoint and place its key in OPENARG_API_KEY; the setup can then be tested with listar_fuentes.

Self-hosting the repository is not a standalone replacement for the hosted service. The code forwards requests to an OpenArg backend, so a local deployment needs a backend configured through BACKEND_URL. Authentication and quota enforcement are handled by that backend rather than by the MCP server process.

Tools and capabilities

• listar_fuentes reports covered open-data portals and the number of datasets in each. It does not consume question quota.
• buscar_datasets searches more than 30,000 catalog entries by Spanish text and can filter by portal. Results include queryable table names and official links.
• describir_tabla exposes schema details, row count, covered period, freshness information, and a sample before data retrieval.
• obtener_datos returns CSV rows with selected columns, date ranges, filters, ordering, pagination, and up to 500 rows per request. It supports equality, comparison, substring, and membership filters.
• agregar_datos calculates database-side sums, averages, counts, minimums, or maximums, optionally grouped by up to three columns. It supports date and filter constraints, ranking order, result limits, and survey weighting.
• consultar_datos_publicos handles natural-language questions and attaches the datasets used, official links, warnings, and remaining monthly questions.

Limitations and notes

The standard allowance is 200 data-mode requests per month and 10 answer-mode questions per month. Data mode is limited to 30 requests per minute and 500 returned rows per request; answer mode is limited to two questions per minute. Quotas reset on the first day of each month at 00:00 UTC. When a quota is exhausted, available credits may be used; otherwise the API returns HTTP 402 with the reset date.

An answer-mode question consumes one question except when it errors, times out, or repeats the same question within five minutes. Data-mode tools do not consume answer questions. The answers mode uses a language model, so users should check the linked official source before publishing a figure. Freshness metadata distinguishes when a table was read from the date of the underlying data; a recently read table is not necessarily current.

OpenArg — Argentina public data server tools and outputs are in Spanish, although the assistant may be prompted in another language. The repository is licensed under MIT.