Community showcase: KenyaADK — first ADK implementation for East African civic data

[gabrielmahia]

Community showcase — KenyaADK: the first Google ADK implementation serving East African public civic data.

What was built

Agents:

• drought_agent — NDMA drought phases for all 47 Kenya counties
• budget_agent — Controller of Budget county development fund absorption
• parliament_agent — MP records, bills, CDF from Kenya's 13th Parliament
• rights_agent — Constitution of Kenya 2010 in English and Kiswahili

Root orchestrator routes queries to the right specialist and coordinates multi-agent responses.

Example interaction

User: "Which counties in northern Kenya are in drought crisis and can I send SMS alerts to farmers?"

KenyaADK:
  → drought_agent: Marsabit (Phase 3), Turkana (Phase 4), Wajir (Phase 3)
  → MCP tool: sms_send (via mpesa-mcp / Africa's Talking)
  → "3 counties in crisis. Ready to send alerts to 1,247 registered farmers — confirm?"

The complete protocol stack

KenyaADK demonstrates all three Google AI protocols working together:

• ADK — agent orchestration (this repo)
• MCP — tool access
• A2A — agent-to-agent

Data

Built on published civic datasets with academic DOIs:

• Kaggle: 10.34740/kaggle/dsv/15473045
• HuggingFace: 10.57967/hf/8223

Stack

• google-adk>=1.31.0 (Apache 2.0)
• Gemini 2.0 Flash (free tier at aistudio.google.com)
• Python, pandas, FastAPI (via ADK server)

Would be glad to contribute a tutorial or cookbook entry for ADK with civic/public sector use cases — especially around emerging markets and East Africa.