OpenArg — Argentina public data

org.openargv1.0.0更新于 Sep 30, 2026

Argentina's official public data (INDEC, BCRA, 38 portals): search, read tables, cited answers.

已验证Streamable HTTP可网页运行Data & AnalyticsWeb Search & Scraping

概览

AI 生成的概览

让助手搜索并读取阿根廷官方公共数据集，涵盖 INDEC、BCRA 及 38 个政府门户，并附来源引用。

功能
将助手连接到来自 38 个政府门户的 33,000 多个阿根廷官方公共数据集目录，包括 INDEC 的人口普查、家庭调查、贫困、GDP 和 CPI 数据，以及中央银行、各部委、省份和国会的数据。工具包括 buscar_datasets 用于目录搜索，describir_tabla 用于查看列类型、行数和样本，obtener_datos 用于按列、日期和等值条件筛选 CSV 行，listar_fuentes 用于查看门户覆盖情况，consultar_datos_publicos 用于带来源的自然语言回答。所有工具均为只读，并返回数据集标题、门户和官方文件链接。
适用场景
当助手需要阿根廷官方统计数据或公共数据（如通胀、贫困、GDP 或央行利率）并应引用来源时适用。若希望助手自行对数据行进行推理，可选择数据模式；若希望由 OpenArg 选取并交叉核对表格，可选择回答模式。
运行要求
远程 Streamable HTTP 端点 OpenArg API 密钥，通过 Authorization 请求头以 Bearer 令牌发送。无需本地运行时；服务器无状态，会将密钥转发给 OpenArg API。
安装前请注意
Authorization 请求头携带密钥，请避免将其放入共享配置或日志。每个密钥有配额限制：数据模式每天 200 次请求，回答模式每天 10 次，配额在 00:00 UTC 重置。回答模式使用语言模型，可能出错，发布数字前请核对所链接的来源。数据反映各机构发布的内容。

安装

在 SourceWeft 中

  1. 打开 控制台中的 OpenArg — Argentina public data，将其添加到工作区。
  2. 为需要使用其工具的对话启用该服务。

Web executable，通过 Streamable HTTP。 远程服务在工作区中配置后即可从网页运行时运行。

其他 MCP 客户端

把它添加到你客户端的 mcpServers 配置中。

{
  "mcpServers": {
    "datos-publicos": {
      "type": "http",
      "url": "https://mcp.openarg.org/mcp"
    }
  }
}

README

OpenArg MCP — Argentina's public data, in your AI assistant

mcp.openarg.org · Español ↓

A remote Model Context Protocol server that connects Claude, Cursor, VS Code or your own agent to 33,000+ official Argentine public datasets from 38 government portals: INDEC (census, household survey, poverty, GDP, CPI), the Central Bank (BCRA), national ministries, provinces, municipalities and Congress.

Every answer comes with its source: the dataset title, the portal and a link to the official file.

  • Endpoint: https://mcp.openarg.org/mcp (Streamable HTTP)
  • Auth: Authorization: Bearer oarg_sk_…. Free key at openarg.org/desarrolladores
  • Cost: free

Quick start

Claude Code

bash
claude mcp add --transport http openarg https://mcp.openarg.org/mcp \  --header "Authorization: Bearer oarg_sk_YOUR_KEY"

Cursor (~/.cursor/mcp.json) and other clients with mcpServers

json
{  "mcpServers": {    "openarg": {      "url": "https://mcp.openarg.org/mcp",      "headers": { "Authorization": "Bearer oarg_sk_YOUR_KEY" }    }  }}

Claude Desktop (through mcp-remote)

json
{  "mcpServers": {    "openarg": {      "command": "npx",      "args": ["-y", "mcp-remote", "https://mcp.openarg.org/mcp",               "--header", "Authorization:${OPENARG_AUTH}"],      "env": { "OPENARG_AUTH": "Bearer oarg_sk_YOUR_KEY" }    }  }}

VS Code (.vscode/mcp.json). The key is asked once and stored encrypted.

json
{  "inputs": [{ "type": "promptString", "id": "openarg-key", "description": "OpenArg key", "password": true }],  "servers": {    "openarg": {      "type": "http",      "url": "https://mcp.openarg.org/mcp",      "headers": { "Authorization": "Bearer ${input:openarg-key}" }    }  }}

Your own agent. Put the key in OPENARG_API_KEY and tell your assistant:

Add the OpenArg MCP server following https://mcp.openarg.org/llms.txt. My key is in OPENARG_API_KEY. Test the connection with listar_fuentes.

Tools

All tools are read-only. Tool names and outputs are in Spanish; ask in any language.

ToolModeWhat it does
buscar_datasetsdataSearch the catalog: title, portal, official link and queryable tables
describir_tabladataColumns and types, row count, period covered, 5-row sample
obtener_datosdataRows as CSV (up to 500): pick columns, date range and equality filters. No SQL
listar_fuentesdataPortals covered and dataset count per portal
consultar_datos_publicosanswersA natural-language question answered by OpenArg, with sources and warnings

Data mode. Your assistant does the reasoning over the rows. It is fast (about 1 s) and allows 200 requests per day.

Answers mode. OpenArg picks the tables, cross-references them and checks the numbers against the source. It allows 10 questions per day.

Limits

Per key
Data mode200 requests/day, 30/min, up to 500 rows per request
Answers mode10 questions/day, 2/min

Quotas reset at 00:00 UTC. The full table is at mcp.openarg.org/limites.html. OpenArg reflects what each agency publishes. The answers mode uses a language model and can be wrong, so check the linked source before publishing a figure.

This repository

This is the code of the server behind mcp.openarg.org:

  • mcp_publico/core.py: pure logic, standard library only.
  • mcp_publico/server.py: MCP SDK and HTTP routes.
  • mcp_publico/site/: the website.

The server is stateless and makes no decisions about quotas or authentication. It forwards the user's key to the OpenArg API, which validates it, applies the quota and answers. Running it yourself only makes sense next to an OpenArg backend (BACKEND_URL). To just use the data, connect to the hosted endpoint above.

It is developed in colossus-lab/openarg_backend (mcp_publico/) and mirrored here.

Español

OpenArg MCP conecta tu asistente de IA a más de 33.000 datasets públicos oficiales de Argentina, de 38 portales: el INDEC (censos, EPH, pobreza, PIB, IPC), el BCRA, ministerios, provincias, municipios y el Congreso. Cada dato viene con su fuente oficial.

  1. Conseguí tu clave gratis en openarg.org/desarrolladores.
  2. Conectá https://mcp.openarg.org/mcp con el header Authorization: Bearer oarg_sk_…. Hay guías para cada cliente en mcp.openarg.org/empezar.html.
  3. Preguntá: "¿Cómo viene la tasa BADLAR en 2026?", "Mostrame el gasto público por ministerio del último ejercicio".

Hay dos modos. El modo datos permite 200 pedidos por día: tu asistente busca y lee las tablas. El modo respuestas permite 10 preguntas por día: OpenArg arma la respuesta y verifica los números.

License

MIT. Un proyecto de Colossus Lab · [email protected]

来源：README.md，提交 bf03984

工具

0
工具元数据尚未被收录。

版本历史

1
  1. v1.0.0最新Sep 30, 2026