On the only Stripe-verified revenue database, the AI category has the lowest median revenue of any category measured.

Seven rungs. The expensive decisions here are all made before any code ships, and every one of them is a decision to not do the thing the market is currently doing.

1. Do not build an AI product

   The AI cohort has the lowest median revenue of the four categories measured across 5,079 Stripe-verified projects — $156/month against $1,170 for marketplaces. Crowding shows up in the median long before it shows up in the discourse. Build in a category with a boring, well-understood job to be done and a clean API surface.

   AI $156 · e-comm $599 · sales $711 · marketplace $1,170

2. Spend nothing on search

   The $12,000 Postiz put into SEO returned, by its own account, minimal traffic, and was discontinued. This is the single clearest use of the $500 in the whole route: not spending it. A budget that size cannot buy attention in 2026, and the channel that eventually worked did not charge for placement.

   $12,000 spent → discontinued · $0 spent → 15%/mo

3. Pick AGPL-3.0, and understand why

   Open source is not the giveaway it reads as. AGPL obliges anyone offering a modified version over a network to publish their changes, so the 6,500 forks are contributors rather than competitors. A permissive licence at the same star count would have funded someone else's hosting business. The licence, not the code, is what is defensible.

4. Ship the CLI before you ship the API docs

   An agent pays for every token it emits. postiz integration list is a handful of tokens; the equivalent JSON request body is hundreds, billed to the operator on every call. The terse interface wins on cost, and cost is the only preference an agent has. This is a genuinely new design constraint and almost nobody is designing for it yet.

5. Publish to the registry and the hub the same week

   Listing an MCP server costs nothing and puts the product in the tool list of every agent that reads that registry. 22.6% of the 31,309 servers there ship no source repository, which means a listing that links a real, starred, auditable repo is in the better quartile of the shelf on day one, for free.

6. Price at $29 and count to 173

   The verified ARPU of the reference business is $31.30 on a $29 entry plan — the upsell is doing almost nothing, and the forecast is therefore a count of accounts, not a blended guess. Five thousand a month is 173 of them. Anything that does not add accounts at $29 is not a growth activity.

   $208,089 ÷ 6,648 = $31.30 · $5,000 ÷ $29 = 173

7. Assume the shelf closes

   Registrations went from 456 a month to 6,673 in a year. Free distribution channels are free during the interval between a standard existing and the incumbents noticing, and that interval is measured in quarters. Treat the registry the way an early advertiser treated cheap inventory: use it hard now, and expect the price to be set by someone else by the time it is obvious.

   456 → 6,673 new servers/mo · 14.6× YoY