mcp-banxico

Problem

AI assistants don’t know what the peso is worth today. Ask them for the exchange rate, inflation or UDIS and they answer with stale training data or, worse, a made-up number. For any financial use in Mexico that’s unacceptable: the official reference is the Bank of Mexico (Banxico), and an answer with no source or date is useless.

Solution

mcp-banxico is a Model Context Protocol server that gives any compatible client (Claude Desktop, Cursor, Cline, Gemini CLI) direct access to Banxico’s Economic Information System (SIE). It’s not a scraper or a third-party mirror: it queries the official source, and every answer includes the cut-off date and the series ID, so it can be audited.

It exposes six read-only tools:

Tool	What it reads
tipo_cambio_usd	USD/MXN exchange rate (FIX)
inflacion_mexico	INPC inflation, annual or monthly
valor_udis	UDI value
tasa_interes_banxico	Target rate and TIIE
reservas_internacionales	International reserves
consultar_serie_sie	Any SIE series by ID and date range

Architecture

There are two ways to use it:

• Locally, with uvx mcp-banxico or pip install mcp-banxico, using the free SIE token. The client talks to the server over stdio (JSON-RPC).
• Public remote endpoint at https://banxico-mcp.duckdns.org/mcp, over streamable-http. No install and no token; it is rate-limited per IP and runs on my own infrastructure (Linux, Nginx and TLS).

{
  "mcpServers": {
    "banxico": {
      "type": "streamable-http",
      "url": "https://banxico-mcp.duckdns.org/mcp"
    }
  }
}

The live page: Banxico Daily

So anyone can see the data without installing anything, I built Banxico Daily (in Spanish) on the same series the MCP server reads. It’s a public page with 22 official indicators (exchange rates for five currencies, target rate, TIIE, CETES, inflation, UDIS and reserves), each with its chart, downloadable CSV history and a daily reading written automatically from the data: how much it moved against the previous business day and over the last month, in plain language.

The data refreshes on its own every business day through a GitHub Action, right after Banxico publishes the FIX. The technical details are in its case study.

Results

• Monthly PyPI downloads growing steadily (live number above).
• Listed in TensorBlock/awesome-mcp-servers.
• Foundation for other projects: Banxico Daily uses its series, and the Gordito agent ships it as a tool.
• A ten-minute uncut video (in Spanish) shows a local model (Qwen 2.5 on Ollama) and Claude Desktop going from “I don’t have real-time access” to answering with the official exchange rate.

Stack

Python 3.10+, official MCP SDK, Banxico SIE REST API, PyPI publishing, Nginx and Let’s Encrypt for the remote endpoint.