Open source
Made with AI.
Here’s exactly how.
A community library of tools, apps and agents built with AI. Every project shows the agents, the stack, the MCP servers and a recipe for rebuilding it, and the open ones are one click from your own GitHub.
- projects
- 248
- projects
- forkable repos
- 61
- forkable repos
- build recipes
- 187
- build recipes
- MCP servers in use
- 38
- MCP servers in use
From the community
48
How do you stop your agent from rewriting files you didn't ask it to touch?
Théo Martin·Help
132
My MCP set-up for design work: Figma, Are.na and Linear
Lena Vogt·MCP
97
Should showcase-only projects rank lower than open source ones?
Noah Fischer·About whatwemade
156
Prompt pattern: make the agent write the failing test before the fix
Priya Raman·Prompts and workflows
Marketplace
Coming later: buy a running project from the maker who built it, with the code, the users and a proper handover.
Your agent can make the listing too.
Paste this prompt into Claude Code, Codex or Cursor. It writes a whatwemade.json and a build recipe from your repo, takes the screenshots and records a short demo. Connect GitHub and we pick it up automatically.
Asset pack prompt
You are preparing this project for whatwemade.app, a community library of things made with AI. Create a folder called /whatwemade in the project root with the following, then stop and show me what you made.
1. whatwemade.json, a manifest with these fields:
title, tagline (max 70 characters), description (2 to 4 short paragraphs, plain language, no marketing words),
type, category, tags (max 5), links (repo, live),
agents: [{ name, role, share }] for every AI tool used, with an honest share of the work in percent,
stack: [frameworks, databases, hosting],
mcp: [{ name, use }] for every MCP server used while building,
build: { time, cost, prompts }.
Read the git history, package files and any agent rules files (AGENTS.md, CLAUDE.md, .cursor/rules) to fill this in. Ask me about anything you can't find instead of guessing.
2. recipe.md, a build recipe: one prompt that someone else can paste into their own agent to build this project from scratch without seeing the code. Write it as instructions to an agent with these sections: what it does, stack, how it fits together, data model, build it in this order, tools to connect, watch out for, and done when. Leave out secrets and anything private. Under 1,000 words.
3. cover.png at 1600×1067. A clean screenshot of the most characteristic screen, centred on a plain background colour taken from the app's own palette. No device frames, no text overlays.
4. Three to six screenshots (shot-1.png, shot-2.png, …) at 1600×1067 showing the main flows in the order a new user meets them.
5. demo.mp4, 20 to 45 seconds, 1600×1067, no sound needed. Use Playwright (or the Playwright MCP server if available) to record a real walk-through: open the app, do the one thing it's for, show the result. Move the cursor at a calm, human speed.
6. logo.svg if the project has one.
Use realistic sample data in every screenshot, never lorem ipsum, and never show real personal data, API keys or tokens.