Describe the tracker or calculator you wish you had, and an AI builds it for you — no code, no developer, no software subscription.
Four prototypes you can try right now — nothing to install · 100 copy-and-paste prompts · a 15-minute setup guide.
From Mike Lee — food futurist, author of Mise, founder of The Future Market.
Click any prototype below — it opens right here in your browser, no sign-up needed. The data is made up; the point is to show what's possible.
Like what you see? Build your own version with your real numbers — your menu, your suppliers, your fields. It's a one-time 15-minute setup, then you copy a prompt and paste it in. Every tool runs entirely in your own browser, so the numbers you enter stay on your computer. New here? Start with the Setup Guide, then browse the 100 Prompts for your corner of food — restaurants, kitchens, CPG, farms, distribution, retail, catering, and more.
A vibecoded tool is real software, and real software can leak. Before your tool touches real data — or anyone besides you — run the five-minute check on the Security page. One copy-paste prompt, a plain-English report, no security degree required.
Beyond the four prototypes, there are ten ready-to-paste prompts for each of ten corners of food — restaurants, kitchens, CPG brands, farms, distribution, food safety, grocery, catering, data, and sustainability. Find the one closest to your work and you've got a starting point.
The tool you just vibecoded is real software, and real software can leak. You don't need to become a security engineer — you need four habits and one prompt. This page is both.
An API key lets anyone who holds it spend money as you. Keys never belong in your code, in a file you share, or in a screenshot. If a key ever ended up in any of those places, assume it's burned — go to the provider's dashboard and create a new one. That's called rotating a key, and it takes two minutes.
"Nobody knows the URL" is not protection. Automated scanners find new sites within hours of going live — not because you're a target, but because they scan everything. Before a tool leaves your laptop:
Many of the prompts on this site build tools that save data "in your browser." That's a feature — your numbers never leave your machine. It's also a limit: the data exists in one browser on one device, and clearing your browsing history can erase it.
If your tool uses software packages (most web apps do), those packages get security fixes all the time — and last year's versions carry known holes that scanners actively look for. Once a month, open your project and ask your AI: "Check this project's dependencies for known vulnerabilities and update what's safe to update." That's the whole habit.
The Five-Minute Check
Open your project in Claude Code — or whatever AI tool built it — paste this, and let it work. It checks the six ways vibecoded apps most often get burned, and reports back in plain English. Run it before the first time you share anything, and again after big changes.
Act as a security reviewer. I built this project with AI assistance and I'm not a professional developer, so explain everything in plain language. Review every file in this project and check for: 1. Exposed secrets — API keys, passwords, tokens, or database URLs written into the code or any saved file. Flag anything like this in frontend code especially, since everything that ships to a browser is public. 2. Unsafe input handling — anywhere user input reaches a database query, file path, shell command, or web page without being validated or escaped (SQL injection, cross-site scripting, path traversal). 3. Missing access control — pages, API routes, or admin functions that should require a login but don't, and anywhere one user could see or change another user's data just by editing a URL or an ID number. 4. Data exposure — personal or business data that gets logged, stored unencrypted, or sent to a third-party service without a good reason. 5. Risky dependencies — outdated or known-vulnerable packages. If this project has a package manager, run its audit command and include the results. 6. Loose configuration — debug mode left on, default passwords, wide-open CORS settings, missing HTTPS, or file uploads with no size or type limits. For every problem you find, give me: - A severity: CRITICAL (fix before anyone uses this), HIGH (fix this week), MEDIUM (fix soon), or LOW - The file and line where it lives - What could actually go wrong, in plain English — what a stranger could see, change, or spend - The exact fix, and an offer to make the change for me Don't soften anything. If something would let a stranger read my data, take over the app, or run up a bill on my accounts, say so directly. Finish with a one-paragraph verdict: is this safe to share publicly, safe only within my team, or should it stay private until the critical fixes are done?
Fix every CRITICAL before anyone else touches the app — paste the findings back in and say "fix the critical items." HIGH gets handled this week, MEDIUM when you can. And if the verdict says keep it private until it's fixed, believe it. Re-run the check after big changes the same way you'd re-check a recipe after swapping an ingredient.
| Recipe | Ingredient Cost | Menu Price | Food Cost % | Status |
|---|
Copy this prompt, open Claude Code in an empty folder, and paste. When it asks clarifying questions, give it your actual dishes, portion sizes, and current invoice prices. Everything you see here will regenerate — styled for your brand, wired to your real data.
Build me a single-page HTML app called "Recipe Costing" as one self-contained index.html file with embedded CSS and JavaScript (no frameworks, no build step, no backend). REQUIREMENTS: - Pre-load 10-15 real menu items from my restaurant. For each recipe: name, menu price, and a list of ingredients with portion sizes (e.g. "Atlantic salmon fillet 8oz"). - Pre-load a flat dictionary of ingredient unit costs at current invoice prices. - Compute food cost % = (sum of ingredient costs / menu price) * 100 for each recipe. - Show a summary row at the top: average food cost %, total menu items, count of items flagged. - Sortable table of all recipes with columns: Recipe, Ingredient Cost, Menu Price, Food Cost %, Status dot (green <28%, yellow 28-32%, red >32%). - Clicking a recipe opens a detail panel showing every ingredient and its unit cost, with a total. - Alert panel at the top listing every recipe above 30% food cost, sorted worst first. - A "Update Ingredient Prices" card with number inputs for every ingredient, rendered and populated on first load (not gated behind a recipe selection). Typing a new price recalculates every recipe that uses it in real time. - Use an editorial design: warm cream background, serif display headlines (DM Sans 900 weight), generous whitespace, no heavy card shadows, divider-based layout. - Mobile-responsive. - All data hardcoded — no backend, no fetch calls. - Interactivity must survive table re-renders: clicking a recipe row must always open the detail panel, even after the table has been re-sorted or recomputed. Make sure every <th> the sort logic touches has whatever child element it expects (e.g. a sort-arrow span), and wrap any DOM cleanup in null-safety so a single missing element doesn't silently halt the init script. FIRST, ask me for: 1. My 10-15 menu items with prices 2. Ingredient lists for each 3. Current unit costs (or I'll give you a rough spreadsheet to infer from) Then build the whole thing in one index.html file.
| Channel | Total Revenue | Total Costs | Gross Margin % | Net Margin % | Trend |
|---|
If you sell across DTC, distributors, retail partners, Amazon, or wholesale — this is the tool to find out which one is real and which is a vanity metric. Copy the prompt, paste it into Claude Code, hand it your sales by channel, and get a working profitability dashboard in under an hour.
Build me a single-page HTML app called "CPG Channel Profitability" as one self-contained index.html file with embedded CSS and JavaScript. Include Chart.js via CDN for line chart visualization. No backend. REQUIREMENTS: - Pre-load 3-5 sales channels I use (e.g. DTC Shopify, UNFI distributor, direct store delivery, Amazon FBA, wholesale). - For each channel, 6 months of monthly data: revenue, COGS, and channel-specific costs. DTC costs: shipping, packaging, returns. Distributor costs: distributor margin, trade spend, damage claims. DSD costs: delivery, slotting fees, demo costs, free-fill. - Compute gross margin % and net margin % per channel over the period. - Summary stats: total revenue, blended net margin, count of channel-months below my target net margin (from step 4 below; if I don't give one, default to 15%). - Table showing each channel's totals, gross %, net %, and a trend arrow (last month vs previous). - Bottom row showing blended total across all channels. - Clicking a channel row opens a detailed monthly breakdown table with every cost line item. - Alert panel listing every channel-month where net margin dropped below my target (or 15% if I didn't specify one). Label the alert with the threshold being used so I know which number drove the alert. - Line chart of net margin % by channel across the 6 months. - Editorial design: cream background, DM Sans 900 weight headlines, highlighter-yellow accent (#556400 ink for text, #FFE054 fill for chips), dividers instead of cards, generous whitespace. - Mobile-responsive. FIRST, ask me for: 1. My channel names and short codes 2. Last 6 months of revenue and COGS per channel 3. The specific cost line items I track per channel 4. Any target margin I'm trying to hit Then build the whole thing in one index.html file.
| Supplier | Status | COA | Allergen | Insurance | Audit Score |
|---|
Built for anyone — co-packers, restaurants, distributors, food safety managers — who needs to know which supplier documents are expiring in the next 30 days without digging through folders. Copy the prompt, hand it your supplier list, and you'll have a traffic-light dashboard before lunch.
Build me a single-page HTML app called "Supplier Document Tracker" as one self-contained index.html file with embedded CSS and JavaScript. No backend, no frameworks.
REQUIREMENTS:
- Pre-load 10-15 suppliers. For each: name, COA (Certificate of Analysis) expiration, allergen statement expiration, insurance certificate expiration, most recent audit score, audit type (SQF/BRC/FSSC 22000/unannounced), optional note.
- Today's date is fixed at the top of the file (so the demo is reproducible). Parse it as local midnight (e.g. `new Date(2026, 4, 18)` — JS months are 0-indexed) and parse every expiration date the same way. Do NOT use `new Date('2026-05-18')` because that parses as UTC midnight and the displayed date will shift by a day in most US timezones.
- For each document: compute status as current (>30 days until expiration), expiring (within 30 days), or expired.
- Summary stats: count of fully compliant suppliers, count with warnings, count with expired documents.
- Table of all suppliers with overall status dot, COA status cell, allergen status cell, insurance status cell, audit score badge (green 90+, yellow 80-89, red <80).
- Clicking a supplier opens a detail panel with each document, its expiration date, and its status label.
- Alert panel listing every expired document and every document expiring within 30 days, sorted expired-first then by soonest expiration.
- Handle null/missing dates cleanly (display "N/A").
- Editorial design: cream background (#FAFAF7), DM Sans 900 display, highlighter-yellow accent (#556400 ink for text, #FFE054 fill for chips), dividers instead of card shadows, status cells as small colored squares with checkmarks/exclamations/x-marks.
- Mobile-responsive.
FIRST, ask me for:
1. My supplier list (names only is fine — I'll give you document dates after)
2. Which documents I track per supplier (COA, allergen, insurance, GFSI audit, organic cert, kosher/halal, etc.)
3. Audit scheme and typical score range
4. Whether I want to track any custom fields (e.g. country of origin, ingredient category)
Then build the whole thing in one index.html file.
Built for diversified farms, market gardeners, and small commercial operations who are tired of pulling field data out of three different notebooks and a whiteboard. Give Claude Code your fields and equipment; get a clean dashboard you can actually use during planting season.
Build me a single-page HTML app called "Farm Operations Dashboard" as one self-contained index.html file with embedded CSS and JavaScript. No backend, no frameworks. REQUIREMENTS: - Pre-load 10-15 fields. For each field: name, acres, 2025 crop, 2026 planned crop, cover crop (and when planted), 2025 yield, soil pH, field type (row/vegetable/hay), optional notes. - Pre-load 5-8 pieces of equipment. For each: name (with make/model), current hours, service-due hours, currently assigned field or status, notes. - Summary stats: total acreage, field count, equipment count. - Field grid (card per field): name with acre count, 2026 crop, cover crop, 2025 yield, soil pH, notes highlighted in yellow if present. - Equipment grid: name, hours / service-due hours, hours remaining until service, color-coded alert (red <50 hours, yellow <200 hours, green otherwise). - Clicking a field opens a detail panel with every attribute including 2025 vs 2026 crop, cover crop planting date, yield, soil pH with "low" badge if below 6.0. - Editorial design: cream background (#FAFAF7), DM Sans 900 display, highlighter-yellow accent (#556400 ink for text, #FFE054 fill for chips), divider-based grid (no heavy cards), generous whitespace. - Mobile-responsive. FIRST, ask me for: 1. My field names, acreages, and crop rotation (last year's crop + this year's planned crop) 2. Cover crop strategy per field (if any) 3. Equipment list with current hour meters and service intervals 4. Any soil test data I want tracked (pH, organic matter, nutrient levels) Then build the whole thing in one index.html file.
Three steps. No coding. If something breaks, paste the error into Claude or ChatGPT — same as you would for any other tech question.
Three short steps. Each one takes a couple of minutes.
Opens the text window where you'll paste commands. Every computer has one built in.
Mac: Press Cmd + Space, type Terminal, hit Enter.
Windows: Click the Start button, type Command Prompt, hit Enter.
A window opens with a line that looks roughly like the picture below. That line is called the prompt. It shows your username and computer name (yours will be different — alex@MacBook-Pro is just an example), then ends with $ on Mac or > on Windows. The blinking cursor right after it is where everything you type or paste goes.
$, paste your command, then hit Enter.To run a command later in this guide, you'll click anywhere in this window, paste the command (Cmd+V on Mac, Ctrl+V on Windows), and press Enter. Leave the window open for now.
Puts Claude Code itself on your computer. One command — no Node.js, no separate downloads.
Copy the line for your computer, paste it into the terminal, hit Enter.
Mac:
curl -fsSL https://claude.ai/install.sh | bash
Windows:
curl -fsSL https://claude.ai/install.cmd -o install.cmd && install.cmd && del install.cmd
Text scrolls by for a few seconds. When it stops, you're installed.
Connects Claude Code to your Anthropic account so it can work.
Type claude into the terminal and hit Enter. A browser tab opens. Sign in or create an Anthropic account. Pick a plan — Claude Pro is the lowest tier that includes Claude Code, $17/month annual or $20/month month-to-month. Plenty to start.
If claude comes back with “command not found,” close the terminal window, open a fresh one, and try again — the installer needs a new window to take effect.
Back in the terminal, the plain text prompt is replaced by the Claude Code interface. If you see something like the picture below, the install worked.
It's the same terminal window as before — Claude Code has just taken it over. You won't type computer commands here anymore; you type ordinary sentences describing what you want, and Claude does the work. That's the finish line. You're ready to build.
Turns a plain-English prompt into a working tool you open in your browser.
recipe-costing. Every tool gets its own folder.
cd, then a space, then drag the folder onto the terminal window. Hit Enter. You're now "inside" that folder.
claude and hit Enter.
Claude Code asks follow-up questions. Answer in your own words. Along the way it also asks permission before it creates files or runs commands — that's normal, not an error. Say yes to things you understand. It builds a file. Open it in your browser — find your project folder, double-click index.html — that's your tool. If something's off, tell it in plain English: "Make the headline bigger." "The margin formula should divide by revenue." Two or three rounds and you have something your team can use.
Coming back next week to change something? Open the terminal, type cd and the folder name, run claude again, and ask for the change. Same conversation, same tool, any time.
Common pitfalls — read these before you start so you don't have to learn them the hard way.
Everything above is all you need to build your first tool. The extras below make your tools better or let you share them with your team. None of them are required to get started.
First, a one-time setup for the extras below.
The deploying and skills steps below use npm and npx, which come bundled with Node.js. You don't need Node.js to install Claude Code or build tools — only for these extras. To get it, go to nodejs.org/en/download and ignore the black box of code near the top — that's the developer route, and you don't need it.
Find the line that reads “Or get a prebuilt Node.js® for ___ running a ___ architecture.” Set its two dropdowns — your operating system (macOS or Windows) and your architecture (ARM64 or x64) — then click the green Installer button (.pkg on Mac, .msi on Windows; not the Binary file) and run it. You only do this once, ever.
On a Mac: Open the Apple menu (top-left corner of your screen) and click About This Mac. If the Chip line says Apple M1, M2, M3, or M4, choose ARM64. If it says Intel, choose x64. Any Mac bought since late 2020 is almost certainly ARM64.
On Windows: Press the Start button, type About your PC, press Enter, and find the line labeled “System type.” If it mentions ARM, choose ARM64. Otherwise — which is nearly every PC — choose x64.
Right now your tool lives on your computer. Only you can see it. If you want your sous chef, your field manager, or your business partner to use it from their own device, you need to put it online. That's called "deploying."
Vercel is a free service that takes the file Claude Code built and puts it at a real web address (like yourname.vercel.app) that anyone can visit. Think of it as free web hosting with zero maintenance.
Sign up at vercel.com (free), then paste these commands into your terminal:
npm install -g vercel cd your-project-folder vercel --prod
It will ask you a couple of questions the first time. After that, vercel --prod deploys in about ten seconds. Your tool is now live at a URL you can text to your team.
Skills are free add-ons that teach Claude Code how to do specific things better. Without them, Claude Code can still build your tool. With them, the tool will look more polished and follow better design practices.
These are the four I used to build this site. Paste each line into your terminal:
npx skills add nextlevelbuilder/ui-ux-pro-max-skill@ui-ux-pro-max -g
npx skills add anthropics/skills@frontend-design -g
npx skills add giuseppe-trisciuoglio/developer-kit@shadcn-ui -g
npx skills add vercel-labs/agent-skills@web-design-guidelines -g
Plugins give Claude Code extra abilities. The most useful one lets it open a real web browser, click buttons, and take screenshots of your tool to make sure everything works. Without it, Claude Code builds the tool but can't see what it looks like. With it, Claude Code can QA its own work.
Browse the 100 Prompts page for 100 ideas specific to the food industry, each with a ready-to-paste prompt.
Paste any of these into Claude Code. They're starting points — short paragraphs you'll refine as Claude asks questions. You'll go back and forth, refining. By version two or three, you'll have something genuinely useful. Every one handles paperwork, tracking, or math. None of them replace a human judgment call. New? Five prompts are tagged Start here — they're the gentlest first builds.