Build with AI
Your AI, building your backend.
Connect the assistant you already pay for to a MINT project. Describe the backend you need in plain words — it plans the models with you and builds them, with their API and admin screens. Then ask it questions about your data.
How it works
From a sentence to software, in four moves.
You describe
In plain words, in the AI chat you already use.
AI plans
It looks at what you have and proposes models to check.
Owner · name, phone
Dog · name, breed → Owner
Walk · dog → Dog, date, paid
MINT builds
With the builder’s checks — and only within your role.
3 models created
Tables, forms, pages
Sidebar updated
Your team uses it
Right away. Change it again tomorrow by asking again.
Ask anything
Things people ask on an ordinary day.
Build, change, or just ask — the same conversation does all three.
Make a booking system with rooms, guests and payments.
Add a “VIP” flag to clients and a filter for it.
Which invoices are overdue by more than 30 days?
Put this month’s revenue and new clients on the dashboard.
Only show salaries to people with the finance role.
Draft a home page and an about page for our website.
Analyze your data
Ask your data anything — with any AI.
Give an assistant a read-only key and it becomes your analyst: counts, totals, breakdowns and lists from your live records, in plain words. It reads only what your role can view, in one project, and never changes a thing.
- Claude, ChatGPT, Cursor or anything that speaks MCP
- No exports, no SQL, no BI tool — always the live records
- Read-only keys with expiry; revoke one at once
Works with
Bring the assistant you like.
MINT speaks MCP, the open protocol AI assistants use to work with other tools. Your project’s Connect AI page shows each set-up with your key already filled in.
Claude
Add a custom connector with your project’s URL — on the web or desktop.
ChatGPT
Turn on developer mode and create a connector with the URL.
Claude Code
One command in your terminal adds MINT as an MCP server.
Cursor
Add the server to its MCP settings; the tools appear in Agent mode.
Anything MCP
Any client that speaks Streamable HTTP, with your key as a header.
Project keys
You decide what it may do.
Make a key per assistant, name it after where it’s used, and choose:
Can build
On: it creates and changes models and pages. Off: it can only look and check plans.
Can read records
Answer questions from your data — only pages you can view, never changing anything.
Expires
Never, or after 7, 30 or 90 days, or a year.
Safe by design
Powerful, never unsupervised.
It acts as you
A key can never do more than your role allows, and stops if you lose access.
Same checks as the builder
Everything it builds goes through the same validation as building by hand.
One key, one project
Each key belongs to one project. See when it was last used; revoke it at once.
Nothing hidden
What the AI changes lands in history next to everyone else’s changes.
Your AI is ready. Give it something to build.
Join the waitlist and we’ll open a workspace for your team — set up for what you want to build.