Distribution advice for developer tools usually assumes you have something I didn't: a network, a budget, or a time zone that overlaps with San Francisco. Building FileMayor from South Africa, I had none of those. What I had was a product whose best demo was seven lines of terminal output, and a decision to treat distribution as a design problem instead of a shouting problem.
The channel was the runtime
The single most important distribution decision was boring: npm. Not a waitlist, not a landing page with a demo video — the package registry developers already trust, with the runtime they already have. The distance between "heard about it" and "ran it on my own Downloads folder" is one command, and that command needs nothing installed but Node.
# thirty seconds later it's organising *your* files, not a demo's
Every alternative added a step — download a binary, create an account, grant cloud access. Each step is a cliff most curious people fall off. npm meant the trial was the install, and because the doctrine keeps everything local, trying it requires zero trust decisions about my servers. There are none.
Real output is the only marketing asset
Early on I made mocked-up screenshots. They performed terribly, and they deserved to. The post that first moved the install graph was the actual CLI, uncropped: 1,248 files scanned in 1.4 seconds, a health score of 22/100, a concrete plan — 63 screenshots, 14 duplicate installers, 27 old archives — then apply, then undo. No mocks, runs offline.
The before/after of a chaotic Downloads folder became the most shareable asset we have, because everyone's Downloads folder is a private shame. Showing the undo at the end mattered as much as the cleanup — the audience for a file tool is people who've been burned by one.
Build the launch kit once, spend it everywhere
Solo founders die by re-making assets. So the FileMayor launch kit is one design system with three surfaces: a 16-slide pitch deck for investors and engineers, five social carousels — thirty-five slides at 1080×1350 that work on LinkedIn, X, and Instagram — and a 29-page brand book so every future asset matches without re-deciding anything. One aesthetic, warm cream and terminal green, everywhere. A week of design bought me months of never opening a blank canvas before a post.
The five carousels were aimed, not sprayed: a manifesto piece for the mixed crowd, a five-verbs walkthrough for developers, a founder story — solo, from South Africa, two years on the engine — the Downloads before/after for shareability, and one on MCP setup for AI builders. Which brings me to the audience that actually converted.
The AI-builder wave, ridden honestly
FileMayor ships as an MCP server, which means Claude and other agents can drive it — safely, through the same six layers as everyone else. The "your AI can suggest, FileMayor can act" framing landed hardest with people building agent workflows, because they'd all hit the same wall: models are great at proposing file operations and terrifying at executing them. We didn't bolt AI onto a file tool for the trend; we built the missing actuator and told that audience exactly where the gate is. Highest-virality segment by far, and the one that files the best bug reports.
What I'd tell the zero-install version of me
- Pick the channel with zero new trust decisions. npm + local-first meant nobody had to believe my privacy policy — there's nothing to believe.
- Ship the real output. Uncropped terminal beats polished mockup, every single time. If the real output isn't shareable, that's a product problem, not a marketing one.
- Make assets as a system, not as posts. Deck, carousels, brand book — one week, one aesthetic, reused for months.
- Name your niche audience and build them a door. For us: AI builders, and the door was MCP.
- +700 a week compounds. Nothing on this list is a spike. Spikes made the graph twitch; defaults made it climb.
25,000 installs isn't a big number by Silicon Valley lights. From a one-person studio in Johannesburg with no budget, it's proof of a different theorem: distribution isn't a department. It's a property of the product, decided at design time — same as security, same as everything else worth doing.