datosabiertos-dashboard

name	datosabiertos-dashboard
description	Builds and launches the Peru open-data Streamlit dashboard (health IPRESS + education enrollment) from datosabiertos.gob.pe via MCP. Use when the user asks for datosabiertos dashboard, panel datos abiertos, CKAN salud educación, or datosabiertos-dashboard skill.
disable-model-invocation	true

datosabiertos-dashboard — Panel Salud y Educación

Fetch live data from the datosabiertos MCP server, embed parsed data in dashboard.py, and launch Streamlit.

Steps

1. Fetch data — call get_health_edu_snapshot(200) on MCP server datosabiertos. Returns health_establishments and education_enrollment with records, fields, total, chart_hints.

2. Parse records — each dataset has records: [{col: val, ...}, ...].

   • Skip rows where key grouping fields are null/empty.
   • Use chart_hints and inspect fields to find columns:
     • Health: Departamento, Tipo / Clasificación
     • Education: Año Matrícula, Departamento\nColegio, Género
   • For charts, aggregate with Counter or pandas value_counts (top 12–15 categories).

3. Overwrite dashboard.py — full file rewrite with:

   • Embedded HEALTH_RECORDS, EDU_RECORDS as Python lists of dicts (from snapshot, max 200 rows each).
   • Embedded FETCHED_AT, HEALTH_META, EDU_META dicts.
   • No runtime API or MCP calls in Streamlit.
   • Layout:
     • Title: Panel Datos Abiertos — Salud y Educación (Perú)
     • Subtitle: Fuente: datosabiertos.gob.pe | Actualizado: <fetched_at>
     • Row 1: barras por departamento (salud) | barras por tipo (salud)
     • Row 2: línea/barras matrícula por año | barras por departamento colegio
     • Row 3: markdown con enlaces a datasets + expanders con preview tabular
     • Footer metrics: totales muestra y totales portal
   • Colors: #C8102E (red), #003087 (navy); st.set_page_config(layout="wide"); Spanish labels; x-axis 45°.

4. Launch Streamlit (background):

.\.venv\Scripts\streamlit run dashboard.py

5. Report — tell the user: Dashboard actualizado y lanzado en http://localhost:8501

Error handling

• If a dataset has errors[key] or empty records, show st.warning("No disponible: <name>") in that panel.
• If all datasets fail, write dashboard.py with only st.error() explaining the API is unreachable; still launch Streamlit.

MCP tools reference

Tool	Use
get_health_edu_snapshot(limit)	Preferred for this skill
datastore_search(resource_id, limit)	Single resource query
search_datasets(q)	Discovery (falls back to package_list filter)
get_dataset(name)	Metadata and resource IDs
list_organizations()	Browse publishers

Notes

• DataStore API may return HTML errors; MCP falls back to CSV download for curated resources.
• Workspace root must be DatosAbiertos/ so .cursor/mcp.json resolves paths.