The value is not any single source but the fact that a question spanning them — employment statistics from INE next to a budget dataset from a city council, on the day the gazette published the rule — becomes one conversation. Semantic search over the catalogue matters here, because Spanish dataset titles rarely match the words you would use.
One MCP server over four Spanish public-data sources: the national open-data catalogue datos.gob.es, the statistics institute INE, the meteorological agency AEMET, and the official state gazette BOE. Instead of four clients and four auth schemes, an agent asks one server.
- Catalogue search and download — `search` filters datasets by title, theme, publisher, format or date, or matches by meaning using embeddings; `get` returns metadata and, with `include_data=true`, parses the CSV or JSON in the same call
- Official statistics — `ine_search` finds statistical operations or lists their tables, `ine_download` pulls the actual series
- Weather — `aemet_list_locations`, `aemet_get_observations` for a station, and `aemet_get_forecast`, which accepts a municipality name such as Madrid rather than a code
- The state gazette — `boe_get_summary` for a day's issue, `boe_get_document` by ID, and `boe_search` for full text over a date range
- Five resource templates (`dataset://{dataset_id}`, `theme://{theme_id}`, `publisher://{publisher_id}`, `format://{format_id}`, `keyword://{keyword}`) and six guided search prompts
Python 3.10 or higher. Clone the repo and install with `make dev`, then run `mcp run server.py`. Only AEMET needs a key — set `AEMET_API_KEY` (free) if you want weather; the other three sources are open. Per-source rate limits and log level are configurable through environment variables. Package version 1.2.0.
One command plus a key — mcp run server.py, then supply credentials