72 Hours with Claude Code + Unreal MCP: Building a Souls-Like Boss Fight

A developer shared his hands-on experience on Reddit: in just 72 hours, he built a fully playable Souls-like boss fight inside Unreal Engine 5.8. Throughout the process, AI handled a significant amount of repetitive work, with Claude Code paired with Unreal MCP directly operating the engine project — serving as the core accelerator of the entire development pipeline.

The real value of this case isn't "AI can make games for you" — it's that it clearly illustrates the actual boundaries of AI-assisted game development today. These tools can dramatically compress timelines, but foundational skills remain indispensable.

What Claude Code + Unreal MCP Actually Did

MCP (Model Context Protocol) allows Claude to directly connect to and operate within an Unreal project, rather than just generating isolated code snippets. According to the developer, Claude Code can inspect the project structure, read and process existing Blueprint logic, wire up weapons, build Boss AI, modify game interactions, and even handle dynamic camera behavior — a feature that sits squarely in the UX layer.

He gave a concrete example of a task: "Here's the Boss, here's the Player, here's the existing logic. Wire up this weapon, check the Blueprint configuration, and switch cameras when the fight starts." Given clear, context-rich instructions like this, Claude can understand and execute them. Compared to the traditional workflow of writing code, switching windows, and manually connecting Blueprint nodes, having AI work directly inside the project eliminates a tremendous amount of tedious, repetitive effort.

This is an entirely different level from simply asking a large model to "generate some C++ or Blueprint pseudocode." The significance of MCP lies in bridging the gap between AI and the engine's runtime environment, enabling the Agent to perceive the real project state and make context-aware modifications.

What is MCP? MCP (Model Context Protocol) is an open protocol standard proposed by Anthropic. It was designed to allow AI models to connect to external tools, data sources, and runtime environments in a structured way — not just through text-based conversation. In the traditional "let AI write code" paradigm, AI can only output static text, and developers must copy, paste, and debug it themselves. MCP allows AI to act as an Agent that directly invokes tool functions, reads and writes files, and queries project state. Unreal MCP specifically implements a set of interfaces for Unreal Engine, enabling Claude to perceive the current state of Blueprint node graphs, establish references between Actors, and modify component parameters. This means AI operates at a far more granular level — not just "generating code snippets" but "executing specific operations on a real project." That contextual awareness is what fundamentally distinguishes it from ordinary code completion tools.

The Complete AI + Handcrafted Hybrid Pipeline

Notably, this project wasn't built with a single tool. It's a complete pipeline that deeply integrates AI tooling with traditional 3D workflows. The developer laid out every stage he actually used:

• 3DAIStudio: For prompting, reference images, and quickly switching between different images/3D models via a node-based workflow
• Tripo P2: Handled initial generation of characters, armor, props, and some environment elements
• Blender: Assembly, sculpting, retopology, UV mapping, baking, optimization, and modular scene construction
• AccuRig + manual cleanup: Rigging and weight painting
• Unreal Engine 5.8: Combat, animation, cloth physics, materials, lighting, fog, and final integration
• Claude Code + Unreal MCP: Coding and project operations directly inside Unreal

The core logic of this combination: 3D generation means the developer doesn't have to hand-craft every asset from scratch, saving significant modeling time; Claude Code + MCP cuts down on the enormous amount of repetitive Unreal operations on the development side. Both legs of the pipeline accelerate simultaneously — which is how it all compressed into 72 hours.

What AI Handled vs. What the Human Retained

The pipeline reveals a clear division of labor: AI excels at "generating first drafts" and "executing well-defined instructions" — such as producing base models and wiring up existing logic. Meanwhile, quality control, topology optimization, rigging cleanup, and final integration — tasks that require judgment — were all handled by the developer personally. AI didn't replace any professional stage of the process; it was embedded within each stage to accelerate it.

AI Is Not a Shortcut for Zero-Skill Game Development

The developer's most valuable insight is also his most grounded take on AI's capabilities. He explicitly stated: don't treat this as magic that lets you make games without foundational skills.

His reasoning is straightforward — if MCP steers the project in the wrong direction, you need to know Unreal, Blueprints, 3D, and the structure of your own game well enough to catch the problem and fix it. Otherwise, you risk ending up in a situation where cleaning up the Agent's mistakes costs more time than it saved you.

This observation cuts to a common pain point in AI Agent-assisted development. The Agent's output needs to be verified, and verification requires the user to have the professional judgment to distinguish right from wrong. AI is great at accelerating progress along known paths — but when it goes off course, only someone who knows their craft can catch it in time.

So his framing is accurate: this is a genuinely useful tool that can significantly accelerate your existing workflow once you already understand how that workflow is supposed to work. The speed gains are real — but they're built on a foundation of expertise the user already possesses.

Why Blueprints Require Human Judgment Blueprint is Unreal Engine's visual scripting system that implements game logic through connected nodes, without requiring direct C++ code. For newcomers, Blueprints lower the barrier to entry. But for an AI Agent, a Blueprint graph is structured directed graph data — the node connections, execution flow, and data flow all carry strict semantic meaning. When an Agent modifies a Blueprint, a single incorrect pin connection can cause a logic break or a runtime crash, and the resulting symptoms are often far from obvious. This is exactly why the developer emphasizes "you must understand Blueprint structure" — recognizing whether an Agent's Blueprint modifications are correct requires the user to mentally "run" the node graph. That's a cognitive skill that requires dedicated training, and it's something AI cannot replace.

Takeaways for Indie Developers

For indie developers and small teams, this case offers a replicable way of thinking: don't chase "fully automatic AI game generation." Instead, precisely insert AI tools into the most time-consuming and repetitive stages of your pipeline. 3D generation solves the asset bottleneck; Claude Code + MCP solves the repetitive labor inside the engine; and the human stays focused on design decisions and quality gatekeeping.

Building a playable Souls-like boss fight in 72 hours would have been almost unimaginable just a few years ago. What this reflects is that AI tooling is evolving from "helping you write code" to "directly operating professional software projects" — a genuinely new phase. The maturation of protocols like MCP has given Agents the ability to truly work within complex engineering environments.

But the technical barrier hasn't disappeared — it has simply shifted. It moved from "manual execution" to "judgment and correction." How well you can wield these tools depends entirely on how deeply you understand the underlying workflow.