Claude Code × Unreal Engine 5.8 MCP — A Real Test on 5 Blueprint Challenges
TL;DR · What you'll learn
- 1 Claude Code is hooked into Unreal Engine 5.8's new AI MCP plugin and put through five Blueprint challenges. Can it produce working gameplay systems, wire Blueprint nodes correctly, follow instructions, and ultimately save developer time?
- 2 The test set: build working gameplay systems, connect Blueprint nodes correctly, follow instructions, and most importantly save real developer time. The grader is Smart Poly, a serious Unreal Engine creator.
- 3 On the health system test, the math was correct but a single wiring mistake hooked the Select node into the Print String for subtract/add health. One reconnect later, everything ran clean.
- 4 Damage box decremented 75 → 50 → 25 → 0, health box incremented 25 → 50 → 75 → 100 — exactly as expected. The math was right; only the Print String wiring needed a fix.
- 5 The ragdoll system also came together. A Set Timer by Function delays then calls Recover from Ragdoll — physics off, location and rotation reset, collision re-enabled, ragdoll flag back to false. All auto-generated.
- 6 Final scores across the five challenges: Hit Actor 8–9/10, Soccer Ball 4/10, Hat 8–9/10, Door 6–7/10, Damage Boxes 6–7/10. Overall: a pretty decent job, with prompt precision driving outcome quality.
- 7 Verdict: if you have basic Blueprint knowledge, you can fix Claude's small wiring slips yourself, saving credits and time. Claude Code via Unreal MCP is firmly in the 'practical for real work' zone now.
Read as slides
3 slides total
The test setup — Claude Code wired into Unreal 5.8 via MCP
Smart Poly hooks Claude Code into an Unreal Engine 5.8 project via the new AI MCP plugin and runs it through five Blueprint challenges. The questions are pointed: can it build working gameplay systems, connect Blueprint nodes properly, follow instructions, and — most importantly — save real developer time?
MCP gives Claude Code access to the project's file structure and Blueprint assets, so the test really is 'hand the engineering to the AI.' Smart Poly is a credible grader with strong UE tutorial credentials, so the assessment isn't soft — it's a real-result review.
Health system and ragdoll — perfect math, almost-perfect wiring
On the health system test, the underlying math came out perfect — clamp at zero on subtract, cap at the max on add. The single issue was a misuse of a Select node: it got wired into the Print String for subtract/add health, so the displayed number disagreed with the actual value because Select is a pure function and was being called every time. One reconnect later, everything matched: damage box 75 → 50 → 25 → 0, health box 25 → 50 → 75 → 100.
The ragdoll system was more architectural, but Claude built it cleanly. A Set Timer by Function fires after a delay and calls Recover from Ragdoll, which stops the physics simulation, resets the character's location and rotation, re-enables collision, and flips the ragdoll boolean back to false. The Blueprint-architecture intuition was clearly there.
Five-challenge scores and where Claude Code via MCP actually sits
The final scores across the five tests: Hit Actor 8–9/10, Soccer Ball 4/10, Hat 8–9/10, Door 6–7/10, Damage Boxes 6–7/10. Soccer Ball was the weak spot, but overall Smart Poly calls it 'a pretty decent job.'
The practical takeaway is the important part. Prompt precision drives outcome quality — be specific and results improve. And if you have basic Blueprint knowledge, you can fix Claude's small wiring mistakes by hand rather than burning more credits re-prompting. The combination of Claude Code and the Unreal MCP plugin has moved from 'interesting demo' into the 'practical for real game work' zone.
Editor's Take
The result across five Blueprint challenges — 'the math is right but one wire is off' — accurately mirrors AI-assisted development today. The logic holds, but small connection errors remain, so value depends on a human with basic knowledge fixing them by hand. The realistic sweet spot is not full automation but a division of labor — delegate 80%, finish the 20% — where knowing when to intervene matters more than the average score.
Source
Smart Poly
I Tested Claude Inside Unreal Engine 5.8...
This article auto-summarizes the YouTube video's transcript with Claude. Please refer to the original video for nuance and exact wording.
Watch on YouTube →Related
3 articles
Eric Tech
The Complete OpenCode Tutorial: Keep Claude Code's Playbook (Skills, MCP, agents.md) While Freely Choosing Your Model
Coding agents tend to lock you into one provider and model, but OpenCode lets you freely choose between existing subscriptions, free models, API-key models, or local models -- while keeping the features popularized by Claude Code and Codex.
IBM Technology
MCP vs. Skills: IBM Breaks Down When to Use Each for Building AI Agents
Getting the right answer from an LLM requires the right context -- giving it just a role and task is prompt engineering, while adding extra information is context engineering.
Griffin Wooldridge
The Mobbin MCP Gives Claude Code Access to 600,000+ Real UI Design References
Claude Code can now pull from over 600,000 real UI screens -- onboarding flows, paywalls, settings pages -- before it designs anything.