Loading page…
Spotted mid-flight that Cursor developers had nowhere central to find “rules,” hacked together a directory site in 3 hours after landing, and rode X + Hacker News + organic YouTuber coverage to $34-35K/mo at 99.8% margin — on 3 hours of maintenance a month.
“I thought directory sites died long before 2025” — Pontus disproved it with a 3-hour weekend project: seizing the un-aggregated blank space of Cursor rules and riding polished design + build in public to $34K/mo, 99.8% margin, and 3 hours of maintenance a month.
Fewer bars = easier, cheaper, or faster for an AI-assisted solo builder. Editorial judgments based on the case details.
[Video] cursor.directory is the directory site where Cursor developers go to find and share “rules.” Founders are Pontus and his close friend Victor (design co-founder); their day job is bootstrapping their startup midday, and they've also shipped open-source projects like Languin. This is a Starter Story (Pat Walls) interview. $34-35K/mo, 99.8% margin.
The first wave was not customers, it was developers, and they came from a following that already existed. Pontus and Victor had spent two years bootstrapping midday in public on X, so the launch post for cursor.directory landed on an audience of developers rather than on an empty feed — one post pulled roughly a million impressions, and there were several posts, not one. He also posted to Hacker News and the site reached the front page. Then YouTubers started putting the directory into their Cursor videos without being asked. The supply side was seeded by hand before any of that: a hard-coded JSON file of rules he had dug out of GitHub gists and forum threads, enough to make the site worth clicking around on day one. Design carried the rest. His argument is that in a world where every AI-built site looks the same, polish is the only thing left that makes a link worth resharing — and the reshares are what turned a weekend project into 2.2 million unique visitors and 49,000 registered accounts.
Revenue arrived after the traffic, which is the only order that works for a directory. The site was free at launch and stayed free; money entered through three paid surfaces layered on top of an audience that was already there — companies paying to list job ads, paid featured placement for rules and MCP servers, and a generator that takes an uploaded package.json and hands back cursor rules tailored to that dependency list. In this category the first cheque is nearly always a job listing, because a company hiring Cursor-fluent engineers will pay a few hundred dollars to sit in front of an audience that is, by definition, exactly that. Featured slots follow once vendors can see the traffic, and they price against attention rather than clicks. The cost structure explains why none of it had to be rushed: operating costs run about $525 a month against $34,000–35,000 in monthly revenue, a 99.8% gross margin, so the site could have run unmonetised for a year without hurting. Payments go through Polar as merchant of record, which matters for a European company.
The best living proof that “directory sites aren't dead — you just have to catch the right ecosystem wave”: cursor.directory hit the Cursor-rules blank space plus the MCP second wave, landing 2.2M visitors and $34K/mo at ultra-high margin on minimal cost. Thematically adjacent to the generic “AI directory site” (ai-directory-site) concept already on this radar, but cursor.directory is a concrete product + real founders (Pontus/Victor, day job midday) + real numbers — a benchmark instance of that thesis, not a duplicate. The most reusable part for developers: stalk an exploding developer/AI-tool ecosystem for the “information scattered, nobody aggregating” crack, launch in 3 hours, and self-propagate via build in public. The hard parts are picking the right ecosystem wave and early distribution (you need an existing X audience or Hacker News luck). Best fit for the overseas developer ecosystem.
Paste into Claude Code / Codex and get a working version of this product end-to-end.
Two production-grade prompts per idea — a full build spec and an SEO growth plan — plus 10 tailored build specs a week. Members only.
Join the Unicorn Club — $5/moCancel anytime, in one clickEvery claim traces back to a listed source — open any of them and check the number yourself. Timestamps jump to the second each figure is said.
Members read the exact words said here
0:18 · hear it in the videoThe exact line behind each figure is in the Unicorn Club. Read the quote, then jump to the second it's said. Join the Unicorn Club
Data credibility: Founder Pontus self-reported on the Starter Story podcast: 49,000 registered users, 2.2M unique visitors, 99.8% margin, ~$500/mo in operating costs, just 3 hours of maintenance per month. All revenue figures are founder-reported or estimated by the video creator — unaudited.
Three loops started turning and only the first one took effort. Building in public was the starter: shipping openly and keeping the whole project open source meant every improvement was itself a post, and the posts brought people who then submitted rules as GitHub pull requests. That is loop two — contributors supply the inventory, which is why maintenance is three hours a month instead of a full-time content job. Accounts made it self-sorting: 49,000 registered users post their own rules and vote on each other's, so the popular list stays current without an editor. The real turning point came from outside. When MCP appeared some months after launch, they built a searchable MCP hub on the same site, and that produced a second adoption wave no amount of marketing could have bought. The lesson underneath is that a directory does not compound on its own — it compounds because it sits on an ecosystem that keeps generating new things worth cataloguing, and an exact-match domain quietly collects the search traffic that ecosystem creates.
This project did not have one, and understanding why is more useful than pretending it did. The audience, the design system, the reusable components and the working shorthand all came out of two years of bootstrapping midday. Victor opened Figma while Pontus started a fresh Next.js project, and something was clickable on Vercel within half an hour because neither of them had to learn anything first. The quiet stretch was paid for in advance, on a different project. For anyone copying the model without that head start, the structure absorbs silence unusually well: at $525 a month and three hours of maintenance, a directory that gets no traction costs almost nothing to keep alive while it waits for its ecosystem event — and this site's own second wave arrived months after launch, from MCP, not from anything the founders pushed. So the judgement call is about the wave, not the site. If the ecosystem is still producing scattered artefacts nobody aggregates, keep it running; if the tool ships an official registry, the crack has closed.
[Video] On a 3-hour flight to France, Pontus watched Cursor videos and noticed “rules” — a recurring pattern with nowhere on the internet to find them in one place. On landing he immediately called Victor: Figma design + a fresh Next.js project + rules hardcoded into JSON — clickable on Vercel in half an hour, live in 3 hours, then bought the cursor.directory domain and kept iterating. Years of working together — “we know each other's DNA” — was the key to the speed.
[Video] Launched straight into build in public on X (a single post pulled ~1M impressions off an existing developer following), hit the Hacker News front page, and YouTubers covered it unprompted. 49,000 signups, 2.2M visitors. Monetization: paid job ads from companies, paid featured slots (rules/MCP), and upload-your-package.json custom cursor-rules generation. When MCP emerged, a searchable MCP hub brought a second wave of adoption. Stack: Next.js + TS + shadcn + Resend + Vercel + polar for payments; ~$500/mo in costs, 3 hours of maintenance per month.
[AI] Disproves “directory sites are dead”: the key was hitting the “information scattered, nobody aggregating” crack inside the exploding Cursor/MCP ecosystem. The reusable model = watch for information gaps in emerging developer/AI-tool ecosystems → launch in 3 hours → self-propagate via build in public. Thematically adjacent to the generic “AI directory site” concept on this radar, but it's a concrete benchmark case — not a duplicate. Risks: heavy dependence on a single ecosystem's momentum, easy to copy, and you need your own distribution audience. Best fit for the overseas developer ecosystem.
That was one of the 50 breakdowns open to everyone. The Unicorn Club opens the other 1,285 at the same depth — plus pain-point filters and 10 build specs a week.
Join the Unicorn Club — $5/moCancel anytime, in one click