Build with AI

Give your AI assistant the real Zung API

Coding assistants guess endpoints they can't see. Point Claude, Cursor, Copilot or any agent at these files and it builds against the endpoints that actually exist — and knows which ones are still planned.

llms.txt llms-full.txt openapi.json

llms.txt

curl https://mfidoc.com/llms.txt
# mfidoc — Zung.ai MFI platform docs
> Two REST APIs: the Banking API … and the
> Member Mobile API … Items marked [planned]
> are not live — never generate code that
> calls them.

## Banking API
- [GET /members — List members](…)
- [POST /loans/{id}/repayment — Record a repayment](…)
…

llms.txt

A short map of every doc page, kit and rule, following the llmstxt.org convention. Small enough to paste into any chat.

llms-full.txt

The same map with every endpoint's parameters inlined, for agents that read a whole file before writing code.

openapi.json

OpenAPI 3.0 for the Banking API (Mobile API here). Import into Postman or Insomnia, or generate a typed client.

All three are generated live from this site's database, so they change the moment a page is edited.

Prompts

Starting points that work

Field collection app

Read https://mfidoc.com/llms-full.txt and build a small offline-first web app for loan officers: list a group's members, capture repayments while offline, and sync them to the Zung Banking API using client_ref so retries never double-post.

PAR dashboard

Using the Zung Banking API described in https://mfidoc.com/openapi.json, build a dashboard that shows PAR by bucket and disbursed vs collected for this year. Poll no more than every 10 minutes.

M-Pesa reconciliation

Write a PHP script that takes a CSV of M-Pesa confirmations and posts each as a loan repayment via the Zung Banking API, using the M-Pesa code as both receipt and client_ref.