Philanthropies sit on decades of invaluable data: grant records, evaluation reports, outcome data, policy documents, and hard-won lessons from running programs.
None of this scales. Each approach requires rebuilding from scratch for every new dataset and every new AI tool.
One server serves every AI client, just like one website serves every browser.
The data owner controls what's exposed and how.
It's an open standard: not locked to any vendor.
Your data never leaves your infrastructure.
Chicago Public Schools AI Guidelines...
To Parent/Teacher Answer Engine
MCP doesn't just make numbers queryable.
It makes qualitative data — policy documents, narrative reports, institutional knowledge — accessible the way we've traditionally only accessed quantitative data.
"What does CPS recommend for parents who want to opt out of AI tools?"
"What are the approved positive uses of GenAI for students?"
"Show me classroom examples for middle school science."
"What does the glossary say about hallucination?"
Every answer is grounded in the actual document. Every answer includes attribution.
A recent convening of 15 organizations surfaced real momentum:
ag@workingpaper.co
(I am really excited about this and want to find good ways to get data out in the world. AND: I don't have a boss anymore who can tell me "no, don't go work on that important thing"! Please don't hesitate to email me.)