500 apps to 2 million: the App Store's history is the MCP economy's future

On July 10, 2008, the App Store launched with 500 apps. By September that year there were 3,000. By January 2009, 15,000. Ten years later, at Apple's own count: 2 million. Apple's own retrospective has the numbers — a four-thousand-times explosion in a decade, and the fastest-growing software distribution surface anyone had ever built.

The web's ad stack didn't survive that jump. A banner sold on impressions, designed for a browser tab, had nowhere to live inside a native app. So a new layer got built specifically for the surface that didn't exist yet — AdMob, incorporated in 2006, betting on a platform shift that hadn't happened yet. By the time Google acquired it for $750 million in 2009, the bet had already paid off: mobile-native ad networks were the only ones that could actually reach an app's users, because they were built for the format apps actually shipped in.

That's the whole pattern, and it's not really about apps. Every platform explosion mints a new monetization layer, because the old one was built for a surface that no longer exists. Web minted display ads. Apps minted mobile ad networks. Search minted search ads. The constant isn't the format — it's that the old layer never survives the jump, and something purpose-built always replaces it.

The same shift is happening again, right now

We're three weeks into indexing the current one. As of today, 80,000+ MCP servers are live across the ecosystem — tools an AI agent calls mid-task, the same way an app called an API. The growth curve rhymes with 2008 closely enough that it's worth saying plainly: this is not a niche developer trend, it's the App Store moment happening to software one layer up the stack.

And the surface has changed shape the same way it did in 2008. An app lived on a home screen a person tapped. An MCP server lives inside whichever client an agent is running in when it reaches for a capability — Claude, ChatGPT, a terminal, a CI pipeline, a Telegram bot. Nobody taps an icon; a model decides which tool wins the call, in milliseconds, with no human watching most of the time. Those clients are the new home screens. Whoever ranks well inside them is the new "featured on the App Store."

What doesn't survive the jump, again

The banner ad couldn't live inside an app because an app has no page to display it on. The same is true one layer further in: an impression-based ad can't live inside a tool call, because there's no screen, and often no human at all — a CI job invoking a tool has nobody to show a banner to, ever. Whatever monetizes this layer has to be native to what a tool call actually is: structured data, returned alongside a real result, paid for only when it's actually used.

That's the shape we built Lulu Ads around, and it's why we built it as middleware at the tool layer instead of a feature inside any one chat product — a chat app is one home screen; the tool layer is every home screen there will ever be, including the ones without a screen at all.

Where this goes

Nobody remembers the ad network that tried to force the web's banner into an app. The ones that won the App Store moment built something native to the surface, early, before the market fully agreed it needed one. That's the bet we're making on the agent economy — three weeks and 80,000 servers into a curve that's still on its steep part.

If you run an MCP server, you're part of the surface this is being built for either way — check your listing. If you want in on the layer that gets built for it: getlulu.dev/publishers.