For the last few years a big part of my work has been pulling data out of DHIS2. Anyone who lives in it knows the routine. Find the right indicator. Find the right org unit. Get the period right. Export. Clean. Repeat.
Last week I tried something different. I built a small tool that just lets you ask.
It is a DHIS2 MCP server. MCP is the protocol that lets AI assistants like Claude connect to real systems instead of guessing. So now I can type something like “show ANC first visit coverage for Dhaka division over the last 12 months” and it finds the indicator, finds the area, pulls the numbers and explains them. No UIDs, no manual export.
It works with any DHIS2 instance, not just mine. You point it at a URL, give it a login, and that is it. Read only by default, so it can look but not touch.
A few things it does:
- Checks the connection and tells you the instance version
- Searches metadata, so you never have to remember a single UID
- Pulls aggregate analytics with normal time ranges like last 12 months
- Reads tracker events and enrollments
I wrote it in Python over a few evenings, with Claude as my coding partner. It is open source under MIT, so anyone working with DHIS2 can use it, break it, or make it better.
This is the part of AI I actually find useful. Not replacing the analyst. Just removing the twenty clicks between a question and an answer.
If you work with DHIS2 or health information systems, I would genuinely like your feedback.
Repo: GitHub - ahasan722/dhis2-mcp-server · GitHub
dhis2 #PublicHealth #HealthInformatics #MCP #OpenSource #HealthData