Claude + Hexels MCP Hands-On: Three Multiplayer Games from a Single Prompt, Total Cost $68

From Gray Blocks to Polished Games: A Turning Point for AI Game Development

If you've been following AI-generated games, you've surely seen those "runs but looks terrible" demos — capsule characters, gray blocks, and entire worlds with a single texture. The code logic works fine, but the visuals are painful.

Recently, overseas creator Adil spent a single afternoon building three multiplayer games with complete art assets — all with zero coding. After publishing them, he woke up to find nearly 4,000 players and 121 remix versions. Even more striking: he showed these three games to Smilegate, the developer behind CrossFire with over 670 million players, and their CEO's reaction was: "The results were much better than I expected."

The core combo behind this: Claude Fable 5 + Hexels MCP. Here's a breakdown of exactly what this workflow does.

Claude + Hexels MCP: Division of Labor Between Code and Visuals

The entire setup takes just 30 seconds in two steps: paste the Hexels MCP connector into Claude and log in, then load a custom Skill (essentially an AI game design team).

Two key concepts need explaining here. MCP (Model Context Protocol) is an open protocol released by Anthropic in late 2024, designed to provide LLMs with a standardized interface for calling external tools. In traditional AI development workflows, models can only output text or code — they can't directly operate external services. MCP changes this — it allows AI models to connect to third-party tools like calling an API, enabling file read/write, database queries, image generation, and more. Hexels MCP is a connector built on this protocol, allowing Claude to directly invoke Hexels' art asset generation, multiplayer hosting, and deployment services while generating game code, creating a complete pipeline from concept to finished product.

Custom Skill is essentially a set of preset system prompts and workflow templates that define the role settings, work steps, and quality standards the AI should follow when executing tasks. In the game development scenario, this Skill shapes the AI into a complete game development team — incorporating the thinking patterns of designers, programmers, artists, and testers. This explains why the AI doesn't jump straight into coding after receiving a request, but instead starts with requirement clarification and design document writing — this behavior pattern is the professional workflow preset in the Skill.

Adil ran a comparison experiment: the same one-line prompt, given to plain Claude versus the Claude + MCP combo. Plain Claude generated a game with perfectly correct logic, but the visuals were nothing but gray geometry. With Hexels MCP added, the game had realistic wooden hull textures, detailed character skins, ocean effects, and complete sound design.

In Adil's words: "This is the dividing line between a tech demo and a game you actually want to play. Fable writes the game, Hexels makes it look real."

It's worth adding some context on Claude Fable 5's capabilities. Fable 5 is a version in the Claude model series specifically optimized for creative coding and game development scenarios. Unlike general-purpose coding assistants, the Fable series has been enhanced during training with deeper understanding of game mechanic design, physics engine logic, real-time interaction systems, and more. It can generate code containing complete game loops (input handling, state updates, render output) and understands core game development concepts like collision detection, particle systems, and network synchronization.

But the visual gap isn't even the biggest breakthrough. The real killer feature is multiplayer deployment — the traditional approach requires hiring backend developers (starting at $50/hour), spending weeks synchronizing player states, and renting servers monthly. Now, friends can join a game by clicking a link, with Hexels automatically handling matchmaking and synchronization — zero additional work.

To appreciate the weight of this breakthrough, you need to understand the traditional technical barriers of multiplayer synchronization. In traditional multiplayer online game development, network synchronization is one of the biggest technical challenges. Core problems developers must solve include: authoritative server architecture design (to prevent cheating), state synchronization strategies (frame sync or state sync), network latency compensation (client-side prediction and server rollback), reconnection mechanisms, and server scaling as player counts grow. A 22-player competitive shooter needs to synchronize hundreds of position, shooting, and damage events per second across all clients. Hexels abstracts all this complex backend infrastructure into a platform service — developers don't need to worry about the underlying implementation, effectively compressing weeks of backend development work to zero.

Three AI-Generated Games Tested: From Pirate Ships to Hand Tracking

Game 1: First-Person Pirate Ship Combat

The prompt was a single sentence: "Build a first-person pirate game where I sail a galleon, fire cannons at enemy ships, and board them for a sword fight on the deck."

The result: a complete pirate game with sailing, cannon fire, and boarding melee combat. Every texture, every sound effect was generated from scratch by the model — realistic wooden hulls, detailed cannons, and character models that actually look like pirates and naval officers.

Game 2: Block World Multiplayer Shooter

The development process for this game is particularly noteworthy. When Adil entered his prompt, the custom Skill didn't start coding immediately. Instead, it began asking questions like a real game studio interviewing a client — inquiring about mechanic details, art direction, level structure, and sound design style. It then generated a complete design document based on the answers, and developed the game from that document.

In the final product, each gun has a unique feel: the sniper rifle fires slowly but deals high damage and has a scope, while the rocket launcher fires rockets with smoke trails that can destroy block structures. The game supports up to 22 players in the same match.

Game 3: Camera Hand-Tracking Fruit Ninja

This was the most impressive one. No controller, no mouse, no touchscreen needed — your camera tracks your fingers, and your fingertips become the blade. The game implements hand tracking, velocity prediction, physics collision, and complete game logic, all running in the browser with nothing to install.

This game is technically remarkable because it coordinates multiple layers of complex technology within the browser. The foundation relies on web-based machine learning frameworks like MediaPipe or TensorFlow.js, which call the camera via WebRTC and use pre-trained hand keypoint detection models to identify the 3D coordinates of 21 hand joints in real time. On top of this, the game needs velocity vector calculations (determining slash direction and force), physics collision detection (intersection between fingertip trajectories and fruit), and complete game state management. All these computations run client-side in the browser via WebGL and JavaScript with no server-side inference needed, placing extremely high demands on code architecture and performance optimization.

As a computer science graduate, Adil admitted this game truly stunned him: "Getting this many complex layers to work together without crashing — that's what amazes me every single time."

Hexels Marketplace Launch: Nearly 4,000 Players with Zero Promotion

After the three games were completed, the Skill asked whether to publish them to the Hexels Marketplace. Adil chose to publish. There's a key distinction here: deployed games are ones you send to others; published games are ones others discover on their own.

The Hexels Marketplace's Remix mechanism borrows from the open-source community's Fork concept and the creative ecosystems of UGC platforms like Roblox. When a game is published, other creators can view the source code, modify gameplay or art assets, and publish their own derivative versions. The value of this mechanism lies in the network effects it creates — every Remix is free promotion for the original, while also enriching the platform's content supply.

The results were remarkable. Adil didn't promote anywhere, didn't run any ads, and woke up to find the block shooter had nearly 4,000 players and 121 remixes — people had opened the game's source code, modified it, and published their own versions. 121 Remix versions means the original game's concept was reinterpreted in 121 different ways. This user-generated content (UGC) flywheel effect is the core logic behind Roblox's multi-billion-dollar valuation.

Adil compared this to the early App Store days in 2008: back then, developers made serious money with nothing more than a flashlight app, for one simple reason — they were among the first to arrive, with zero competition. The Hexels Marketplace is currently at this exact stage.

Cost Breakdown: Three Games for a Total of $68

Adil had Claude pull the complete cost breakdown. Including failed generation attempts, the total cost from development to launch for all three games was $68.

For reference, in traditional indie game development, art outsourcing alone can cost thousands of dollars. A complete animation sequence for a single 2D character typically runs $500–$2,000, and 3D character modeling plus texture painting easily costs several thousand per character. On the server hosting side, a game server capable of supporting dozens of concurrent players typically costs $50–$200 per month. $68 covering all development, art generation, and deployment costs for three games is itself a disruption of the traditional game development cost structure.

The business model is also clear: the Marketplace serves as a free traffic funnel, real player retention data tells you which game is worth betting on, and then you bring validated titles to digital distribution platforms like Steam for scaled monetization. This is essentially an extreme application of Lean Startup methodology in gaming — build a minimum viable product (MVP) at the lowest possible cost, validate demand through real market data, then decide whether to invest more resources in polishing and commercialization.

Smilegate CEO's Verdict: "Much Better Than Expected"

When Adil showed the three games to the Smilegate team, their CEO's reaction was: "The results were much better than I expected, and there will be more use cases in the future." Upon seeing the block shooter, team members said: "This really feels like prototyping," and believed this kind of tool could replicate the content they want to create.

Smilegate is a leading South Korean game developer whose flagship product CrossFire is one of the highest-grossing games globally, with an especially massive player base in the Chinese market (operated by Tencent). Since launching in 2007, the game has accumulated over 670 million registered users, with annual revenue exceeding $1 billion on multiple occasions. In recent years, Smilegate has been actively investing in AI technology, funding multiple AI startups and establishing an internal AI R&D department to explore procedural content generation (PCG) and AI-assisted game design. A major studio with this level of mature commercialization experience showing serious interest in AI-generated games signals that the industry's assessment of this technology has moved from the "proof of concept" stage to the "productivity evaluation" stage.

A studio operating a product with 670 million players seriously evaluating three games generated from a single prompt — regardless of your stance on AI gaming, this signal deserves serious attention.

Final Thoughts: The Window for AI Game Development Won't Stay Open Forever

The core takeaway from this case isn't that "AI can make games" — that's no longer news. The real change is: the entire pipeline from a single sentence to a multiplayer game with art assets that strangers can discover and play has been connected end-to-end, and at a cost individual developers can easily afford.

From a technology evolution perspective, the significance of this "full-pipeline breakthrough" is similar to the leap in the mobile internet era from "being able to write code on a phone" to "App Store + push notifications + in-app purchases." There's a massive gap between a single-point technical breakthrough (AI can write game code) and a complete product pipeline (from concept to distribution to monetization), and the Claude Fable 5 + Hexels MCP + Marketplace combo has bridged that gap for the first time.

Six months from now, games like these may be everywhere, and the Marketplace will be crowded. But right now, this is still a blue ocean window with virtually no competition. The winners of every new platform are always those who showed up first.