A 16-year-old built Grow a Garden in 3 days. It made $12M in its first month.

Roblox paid $1.5B to creators in 2025. On June 8, the rate gets 42% higher for one specific type of game. The platform just shipped its own MCP server so Claude Code can build directly inside Studio. What changed and how to position before the window closes.

In March 2025 a 16-year-old developer who goes by BMWLux opened Roblox Studio and built a farming game in three days. He called it Grow a Garden. Plant a seed, wait for it to grow even while offline, harvest, reinvest. That is the whole loop. He published it March 26 and went back to school.

By April the game had about 1,000 concurrent users. Janzen "Jandel" Madsen from Splitting Point Studios saw it and bought a share. By May it was generating $12 million a month in Robux purchases. By June it hit 16 million concurrent players, beating Fortnite's all-time peak of 15.3 million from December 2020. By August it peaked at 22.3 million during a mock admin-war event with another viral game called Steal a Brainrot. BMWLux still owns about 50%. He turned 17.

I read the Wikipedia entry, the Roblox earnings call, and three Medium posts trying to replicate the playbook. None mention what changed underneath. Three things did, all in the last twelve months. This essay is the breakdown.

If you are reading this and thinking "Roblox is for kids and I should not be here," that assumption is the entire alpha. The 18 to 34 cohort is the fastest growing on the platform, monetizes 50% higher than under-18 users, and Roblox just announced a 42% DevEx bonus for games that serve them. Window opens June 8, 2026.

The rest of this is the proof and the playbook.

Act 1. The economy nobody reads the earnings call for

Forget the marketing posts. Actual numbers from Roblox's official communications and the Q1 2026 earnings call.

Roblox 2025 totals:

Numbers are abstract until you meet the people behind them. Pat Walls flew to Austin to find three.

The DevEx rate. This is the conversion from Robux to actual dollars. It sat at $0.0035 per Robux from 2017 until September 5, 2025, when Roblox raised it to $0.0038. An 8.5% increase. The first rate change in eight years.

Then in April 2026 Roblox went further. Starting June 8, in-game spend by US players age 18 and older in eligible games converts at a much higher rate. The technical framing: today a creator keeps 26.6 cents on every dollar a player spends after platform fees and DevEx conversion. From June 8 that becomes 37.8 cents on every dollar from age-checked US 18+ players in eligible games. That is the "+42% bonus" you keep seeing in headlines: the relative jump from 26.6% to 37.8%. The Incubator and Jumpstart programs that opened in early 2026 had over 8,000 creator applications by Q1.

There is one technical requirement almost nobody reports clearly. To qualify, your game has to use the R15 avatar framework, not the older R6. R15 supports advanced skeletal joints, layered clothing, and modern facial animation. Most games built in the last three years are already R15. If you are on R6, switching is a single toggle in Game Settings, but it requires updating any custom rigs or animations.

Why this matters in plain math. If your game makes 1,000,000 Robux a month from US 18+ players, your DevEx payout under the September 2025 rate is about $3,800. Under the June 2026 bonus it becomes about $5,400 from the same revenue. Same players, same spending, more than 40% more cash to you. Eligible spend covers game passes, Robux subscriptions, select in-game items, and private servers, which are the four monetization streams this essay covers anyway.

The distribution across creators is what the marketing posts skip. Numbers Roblox confirmed in its Q4 2025 shareholder letter:

The shape is brutal. The bottom of the curve makes lunch money. The top of the curve is venture-funded studios with hundreds of employees. But the middle is where this essay lives. The 1,000th best developer on the platform made $1.3M last year, growing 50% year over year. That is not "a few kids got lucky." That is a thousand people clearing well over a million annually from a platform most of the internet thinks is for nine-year-olds. Roblox's CEO confirmed it on the Q1 2026 earnings call: the top 1,000 are growing faster than the top 10. A meritocracy that rewards new entrants right now.

The minimum cashout is 30,000 earned Robux, which under the new rate is $114. The 30% marketplace fee Roblox takes is the same as the App Store. The math is publicly readable.

So that is the economy. Now the part that unlocks it for non-programmers.

Act 2. Roblox shipped its own MCP server. Claude builds Luau inside Studio.

This is the change most coverage missed.

In late 2025 Roblox quietly published an open-source repository at github.com/Roblox/studio-rust-mcp-server. It is a Rust implementation of an MCP server that runs locally and bridges Claude Code (or Claude Desktop, Cursor, Codex CLI, Gemini CLI) directly into a live Roblox Studio session. Roblox built this themselves. First-party tooling.

To make that concrete. Roblox wrote and maintains the bridge that lets Claude Code write Luau scripts into Studio. They did not have to. The MCP standard was created by Anthropic. Roblox shipped their own implementation because they want AI-assisted development on the platform.

Here is the part that closes the argument. On April 15, 2026, Roblox said in their official "Studio is Going Agentic" announcement that 44% of the top 1,000 creators already use Roblox Assistant or third-party AI tools via MCP to plan, build, and test their games. The data is from a one-month window between March 6 and April 7. Roblox's SVP of Engineering, Nick Tornow, named the third-party tools explicitly: Claude, Cursor, and Codex. Forty-four percent of the highest-earning developers on the platform are running this stack right now. The default workflow for people earning $1.3M+ a year.

What this means in practice. You open a Studio project, run claude in a terminal, type "build me a tycoon with a currency system, three upgrade tiers, a leaderboard, and DataStore persistence." Claude writes every line of Luau directly into ServerScriptService, wires up RemoteEvents, sets the right services, tests it by spawning a Play instance. You never open the script editor manually.

The setup is one command:

Then install the companion plugin in Studio, enable HTTP requests in Game Settings, and you are connected.

A developer dryw3st posted a 53-second clip in February. Front-page Roblox game, 100% scripted by Claude Opus 4.6 Extended, four days from blank project to ship. Not a single line touched by a human scripter. His words: "I guided it to complete the entire coding. This is fricken insane."

One anecdote. The 44% figure is the population. A workflow experimental six months ago is now mainstream.

There are now four MCP servers worth knowing.

Start with boshyxd's version. Most tools, longest history, read-only "inspector" mode for safe project exploration. The official Roblox one is best for production reliability but exposes fewer operations.

What Claude actually handles in this stack, based on the Andy.G mining tycoon writeup from April 2026:

The split is right. Code is the hard part if you do not have a programming background. Claude has the code. You have the judgment about what is worth building.

The required CLAUDE.md. Without this file Claude writes vanilla Lua 5.1, or worse, JavaScript. Luau is a strict superset of Lua with type annotations. Claude needs to be told it is in a Roblox environment:

With this file in place, every script Claude produces will run on the first or second playtest. Without it, you will spend an hour debugging type errors before realizing Claude has been writing Node.js handlers.

Here is what a real currency manager looks like when Claude writes it for you. This exact file is sitting in production tycoons right now:

I did not write any of that. Neither did the developer who shipped a working tycoon to the Roblox front page last month. The prompt that produced it was four sentences. The technical barrier the platform used to have, the one that kept smart non-programmers out for fifteen years, is no longer a barrier.

The stack keeps getting deeper. As of April 2026, Roblox's own Studio has gone agentic too. Planning Mode breaks complex prompts into editable task plans before any code is written. A Playtesting Agent runs your game against the plan automatically, reads logs, simulates input, feeds bugs back into the next iteration. Mesh Generation and Procedural Models create 3D assets from text prompts. Available through Roblox Assistant inside Studio or through the MCP server from Claude Code. The same agentic loop you would build yourself in software development now exists natively on the platform.

Act 3. The pattern across $11M-a-month games

If the economy is the why and Claude Code is the how, the last question is what to build. The viral hits of 2025 and 2026 share a structure that is not obvious from playing them.

The five biggest movers right now, with verified monthly revenue from Creator Exchange tracking:

Look at these five games for thirty seconds and the pattern is invisible. Look for an hour and it is everywhere.

Pattern 1. They all have a daily return trigger. Grow a Garden grows crops while you are offline, so logging in tomorrow shows progress. Steal a Brainrot's brainrots generate income passively, so you wake up richer. Brookhaven runs daily login bonuses. The strongest predictor of Creator Rewards payout is daily active users, not total visits. Daily return loops convert one-time players into recurring revenue.

Pattern 2. They ride cultural moments, not technical achievements. Steal a Brainrot is built on the Italian brainrot meme cycle (Tralalero Tralala, Bombardino Crocodilo, Tung Tung Tung Sahur). Mechanics are dead simple. The IP is borrowed from a TikTok trend already going viral. Grow a Garden uses studded textures evocative of 2009-era Roblox, on purpose, because nostalgia among 18-25 year-olds who grew up on the platform is its own engagement engine. Neither game has technically impressive code. Both ride a moment.

Pattern 3. The progression is brutal and visible. Steal a Brainrot has rarities (common, epic, legendary, mythic, Brainrot God), variations on top (diamond, galaxy, bloodrot), rebirths that reset progress for better stats. Grow a Garden has crop mutations. Every session shows numerical proof you are richer than last session. The bigger the number, the more often a player taps "skip wait" or "double income" for Robux.

Pattern 4. Mobile-first by default. Every top-ten game plays well on a budget phone. Steal a Brainrot, Grow a Garden, Rivals, Ink Game, 99 Nights in the Forest. All designed for portrait or landscape touch, all under 100MB. PC is the secondary audience now. If you cannot tap-tap-buy on a four-year-old Android, you are not a viable Roblox game in 2026.

Pattern 5. They get to playable in under a week. BMWLux built the first version of Grow a Garden in three days. Steal a Brainrot's prototype was working in a week. Speed of iteration beats polish. The viral hits all came from teams that shipped fast, broke things in public, let community feedback shape the next update.

The misread. Most coverage frames these games as "anyone can build the next Steal a Brainrot." Wrong. Code is the part Claude handles. The cultural read is what BMWLux had. Picking the right meme cycle, knowing which mechanic feels familiar, knowing when the trend peaks. These are the parts AI cannot do. Claude builds what you describe. Knowing what to describe is the moat.

What this means practically. If you want to ship something that has a real shot, here is the order:

Skip step one and Claude writes you a beautifully architected tycoon nobody plays. Skip step three and you have a single-session game that never makes the Creator Rewards leaderboard.

The four monetization streams that compound

Roblox pays creators four ways. The math compounds when you run all of them at once.

Game Passes. One-time purchases unlocking permanent perks. $1 to $25 each. Conversion on a well-designed pass is 1 to 5% of active players. Math on a mid-tier game with 5,000 DAU and a 3% conversion at 500 Robux (~$5): 150 buyers/day spending 75,000 Robux total. After the 30% platform fee you keep ~52,500/day, ~1.55M/month. DevEx at $0.0038 converts that to about $5,900/month from one pass. Top games run three to five simultaneously.

Developer Products. Consumable purchases players make repeatedly. Currency packs, time-skip boosts, extra lives. The same engaged player might spend 5 to 20 times more here than on Game Passes. The trick is making the purchase moment feel like a natural shortcut, not a paywall. Grow a Garden's "speed up growth" purchases are the canonical example.

Premium Payouts. Passive income from Premium subscribers playing your game. Every minute a Premium member spends in your experience generates automatic Robux with no purchase required. For a game with 30% Premium audience, this can be a third of total revenue.

Private Servers. Monthly subscription a player pays to rent a private instance. Players set their own price (Roblox suggests 200 to 800 Robux/month). Social games like roleplay tycoons and simulator hangouts make 10 to 30% of total revenue from private servers because friend groups want their own space.

Stack all four. The games clearing $1M a year run all four simultaneously. Skipping any of them is leaving money on the table.

What the bonus does to this math. Take the same pass at 1.55M Robux earned/month after fee. Standard DevEx pays you $5,900. At the June 2026 bonus rate for age-checked US 18+ spend in eligible games, that same revenue pays you about $8,400. Same game, same player count, same code. Different rate.

The window. June 8, 2026.

Roblox structured the bonus carefully. Three conditions all have to be true. First, the spending has to come from a US user. Second, that user has to be age-checked as 18 or older through Roblox's verification. Third, your game has to use the R15 avatar framework.

The good news for non-US creators: your location does not matter. Any creator anywhere qualifies as long as your game meets the rules and attracts US 18+ players. The bonus is automatic. No separate application. Eligible payouts get separated at the DevEx cashout stage.

Two kinds of games will dominate the bonus pool. First, games explicitly designed for older audiences that meet Roblox's "novel game" criteria (games with new mechanics, not clones of existing tycoons). Second, existing R15 games that happen to attract a strong 18+ audience. Most independent developers land in the second. A well-designed tycoon, simulator, or roleplay game will naturally attract more 18+ players than a children's obby. The bonus rewards picking that audience deliberately.

What this signals. Roblox is no longer pretending to be a children's platform. On the Q1 2026 earnings call, CEO David Baszucki said the 18 to 34 cohort is the fastest growing user segment, and Roblox is shipping kids-only and select-account modes in June to formally split the audience. The DevEx bonus is the platform putting money where the strategy is. Creators who position now, while the 18+ creator pool is still small, capture the bonus before competition catches up.

This is the actual "window" the marketing posts keep mentioning. Not the platform itself. The specific six to twelve month period where the bonus is in effect and most creators have not adjusted to it yet.

The four mistakes I see in every Medium post about this

This material attracts copycats. Four claims are wrong or misleading. If you see them, the writer has not shipped on the platform.

Claim 1. "Anyone can clone Steal a Brainrot and make millions." No. The IP rights to brainrot characters are licensed. SpyderSammy already sued a Fortnite Creative copy in October 2025. The mechanic is copyable. The cultural moment is not.

Claim 2. "DevEx rate is going up 42% for everyone." No. The bonus is specifically a jump from 26.6% to 37.8% of in-game spend, and it applies only to verified US 18+ player spend in R15 games. Standard rate stays $0.0038 per Robux for everyone else. Read the announcement carefully.

Claim 3. "You need a VPS to deploy." No. Roblox hosts everything: servers, scaling, matchmaking, DataStore replication, global CDN. The "VPS for Roblox" recommendation in some posts is affiliate marketing dressed up as setup advice.

Claim 4. "The top 10 averaged $33.9M each, so this is replicable." The top 10 are studios with employees, infrastructure, licensing deals, and years of compounding visits. The replicable number is the top 1,000 average of $1.3M, growing 50% year over year. Even that requires either luck on a viral concept or three to five years of consistent shipping. The median is $1,575 a year. Set expectations accordingly.

What I would actually do if starting today

If you are a non-programmer with a few hours a week, the 60-day plan looks like this.

Week 1. Install Roblox Studio (free). Install Claude Code on your machine. Add the boshyxd MCP server. Connect them, run get_place_info, make sure Claude can read your project. In Game Settings, confirm your avatar framework is set to R15 (this is the requirement for the June 8 DevEx bonus, and it is on by default for new projects). Build a "hello world" by making a part change color when clicked. Total time about three hours.

Week 2. Pick your concept. Spend one evening reading the Roblox Discover charts, watching the top 50 games at 2x speed, writing down the three mechanics that show up most. Pick a cultural moment you actually understand. Have Claude analyze gaps:

Week 3 and 4. Build the prototype. Currency system, three income tiers, leaderboard, DataStore persistence. Claude writes the code. You handle 3D placement and visual feel. Playable, not polished.

Week 5. Add Game Passes and Developer Products. Starter pack at 50 Robux, VIP at 250, premium at 600. At least one consumable Developer Product (speed boost, double income, skip wait). Claude writes the MarketplaceService integration.

Week 6. Soft launch. Publish. Share in three relevant Discord servers and one TikTok video showing the loop. Watch Day 7 retention. 20%+ means you have something. Under 10% means the loop is broken and no promotion fixes it.

Week 7 and 8. Iterate based on data. The Creator Hub gives you retention curves, revenue per user, drop-off points. Have Claude analyze the data and propose the three highest-impact changes. Ship. Repeat.

This plan is not "make a million dollars in 60 days." It is "have a real game with real players and real (small) revenue, positioned to compound for the next twelve months." The compounding is the part nobody is patient enough for.

The links that matter

Roblox official MCP server:

github.com/Roblox/studio-rust-mcp-server

Community MCP (most popular, most tools): github.com/boshyxd/robloxstudio-mcp

Multi-client MCP (Claude, Cursor, Codex, Gemini): github.com/hope1026/weppy-roblox-mcp

DevEx documentation and current rates: create.roblox.com/docs/production/monetization/engagement-based-payouts

Roblox Incubator and Jumpstart applications:

about.roblox.com (newsroom, April 2026 announcement)

Creator Hub analytics dashboard:

create.roblox.com/dashboard

Studio download (free, Windows and Mac):

create.roblox.com

Honest closing

The real story is not BMWLux or Steal a Brainrot or $33.9M-per-creator headlines. The real story is the structure underneath them.

A platform that paid $1.5B last year. Lifting the effective creator earnings rate from 26.6% to 37.8% on June 8 for the audience growing fastest. First-party AI tooling that compresses the code barrier to a one-line install. A creator economy where 24,500 people have cashed out, the 1,000th best earned $1.3M last year, and that number grew 50% in twelve months.

That is not a kids' platform. That is an economy with a measurable distribution of outcomes and a specific window where positioning beats execution. The window does not stay open forever. The bonus is announced through 2026. The MCP tooling is months old and 44% of the top 1,000 already use it. The 18+ audience is still small enough that even modest games can capture it.

I am building one this weekend. Loom up Friday.

If this saved you a week of research, like, retweet, bookmark.

/ Building in public, reading the data / Embedded post:

Author: Henry (@dryw3st) Post ID: 2024268269608112238 Source: https://x.com/dryw3st/status/2024268269608112238 Reply to: none

Text:

> Today, I'm revealing my Roblox Game FULLY scripted using Claude Opus 4.6 Extended. > > This project took about 4 days to complete, and not a single line of code was touched by an actual scripter. > > In the conversation I had with Claude, I guided it to complete the entire coding, this game started with a Frontend and a Backend - It is now a fully complete frontpage Roblox Game. > > From UI Animations, to Visual effect movements, and UI Icons, it was all Claude. > > The game releases Tomorrow, and I'm expecting it to reach front page on Roblox within a few weeks. > > 300 Likes, and I'll share the FULL conversation, which was used to build the game. > > Claude is incredible, this is genuinely an insane movement for coders enhancing their productivity with AI. > > This is FRICKEN INSANE.

Media:

• video:

Prompts

Claude handles                       Human handles
-------------------------------------------------------------
Game logic (server scripts)          3D placement and decoration
Data structures and modules          Aesthetic judgment
Client scripts and input handling    Visual feedback decisions
Procedural generation                Lighting and materials
DataStore retry patterns             Final polish
RemoteEvent wiring                   Game design decisions
Anti-exploit server validation       What to build in the first place

claude mcp add robloxstudio -- npx -y robloxstudio-mcp@latest
 

Steal a Brainrot     $11M/month     720M hours played    +117% MoM
Grow a Garden        $12M/month     35.3B lifetime visits
Ink Game             $3.2M/month    1.13B hours played   +800% MoM
Rivals               #1 PC revenue  FPS, longest sessions
Brookhaven           $56M lifetime  69B visits, evergreen

# CLAUDE.md (copy this into your project root)
 
## Language
This project uses Luau, not Lua or JavaScript.
Always use type annotations.
 
## Architecture
- Server scripts: ServerScriptService/
- Client scripts: StarterPlayerScripts/
- Shared modules: ReplicatedStorage/Modules/
- Use RemoteEvents for ALL client-server communication
- Use DataStoreService for ALL persistent data, never client-side
 
## Roblox Services
Always use game:GetService("ServiceName")
Never reference services through the workspace directly
 
## Rules
- Server-side validation on every currency transaction
- Never trust the client for any game state
- Use pcall() around all DataStore calls
- Use task.spawn() instead of coroutine.wrap()

Analyze the current Roblox tycoon landscape.
List the top 20 tycoons by current player count.
For each: theme, core mechanic, monetization model.
Identify three themes with strong player demand
and weak existing competition that a solo developer
could ship in two weeks.

-- ServerScriptService/CurrencyManager.server.luau
local Players = game:GetService("Players")
local DataStoreService = game:GetService("DataStoreService")
 
local CurrencyStore = DataStoreService:GetDataStore("PlayerCurrency_v1")
local playerData = {}
 
local function loadPlayerData(player)
    local success, data = pcall(function()
        return CurrencyStore:GetAsync(tostring(player.UserId))
    end)
    if success and data then
        playerData[player.UserId] = data
    else
        playerData[player.UserId] = {coins = 100, incomeRate = 1}
    end
end
 
local function savePlayerData(player)
    local data = playerData[player.UserId]
    if not data then return end
    pcall(function()
        CurrencyStore:SetAsync(tostring(player.UserId), data)
    end)
end
 
task.spawn(function()
    while true do
        task.wait(1)
        for _, player in Players:GetPlayers() do
            local data = playerData[player.UserId]
            if data then
                data.coins += data.incomeRate
            end
        end
    end
end)
 
Players.PlayerAdded:Connect(loadPlayerData)
Players.PlayerRemoving:Connect(savePlayerData)

Roblox/studio-rust-mcp-server      first-party, Rust, official
boshyxd/robloxstudio-mcp           community, 43 tools, most popular
hope1026/weppy-roblox-mcp          newer, multi-client (Claude, Cursor, Codex, Gemini)
code-and-relax/robloxstudio-mcp    fork with 37+ tools, Vector3 fixes, activity logger

1. Pick a cultural moment you genuinely understand
2. Find the simplest mechanic that lets players "collect" or "grow" within it
3. Build the daily return loop first, everything else after
4. Mobile-first UI from the prototype, not as a port later
5. Ship the rough version in week one, iterate in public

Creator payouts (DevEx):       $1.5 billion
Q4 2025 revenue growth YoY:    +43%
Full year bookings 2025:       $6.8B (+55% vs target)
Q2 2025 average DAU:           111.8M (+41% YoY)
Peak concurrent users (Aug):   47.4 million
Developers paid via DevEx:     24,500+

Median creator income (2024):  $1,575/year
Top 1,000 average (2025):      $1,300,000/year (up 50% YoY)
Top 100 average:               $6,000,000/year
Top 10 average:                $33,900,000/year
Highest single game lifetime:  Brookhaven, $56.1M

Links

• github.com/Roblox/studio-rust-mcp-server
• github.com/boshyxd/robloxstudio-mcp
• github.com/hope1026/weppy-roblox-mcp
• create.roblox.com/docs/production/monetization/engagement-based-payouts
• about.roblox.com
• create.roblox.com/dashboard
• create.roblox.com