Community Tutorial for Unreal MCP: Claude Code Team Setup for UEFN is here 🏝️

[mimmo_the_root]

Claude Code Team Setup for UEFN — The AI Development Team That Learns From Its Own Mistakes

If you manage dozens of islands in Unreal Editor for Fortnite (UEFN), you know the headache of generic AI tools losing project context, inventing broken code, or repeating the exact same Verse mistake on island #40 that you already solved on island #12.

Relying on a single, general-purpose AI assistant to juggle coding, bug hunting, documentation, and release planning is inefficient—the agent steps on its own toes and wastes your time.

Built by Mimmo_the_root out of necessity while managing around 50 UEFN projects, the Claude Code Team Setup for UEFN is a free, open-source framework that replaces single-prompt assistants with a specialized team of subagents.

Meet Your Specialized AI Development Squad

Instead of one assistant trying to do everything, five core agents (plus two optional power agents) split the workload with narrow, focused roles:

• project-bootstrap: Maps existing codebases to deduce functional specs and build prioritized bug roadmaps, or asks targeted design questions for brand-new islands. It also delivers retention proposals aimed at the crucial first 5 minutes of gameplay.

• coder: Your dedicated Verse developer. Operates under a strict compile-and-fix loop and never marks a task done if the code fails to build.

• qa-regression: Hunts for regressions and unseen runtime bugs automatically after every playtest session.

• planner-docs: The sole manager of documentation (STATUS.md, ROADMAP.md, BUGS.md), keeping progress current so you never lose context after months away.

• release-gate: An objective release auditor that gives an honest go/no-go verdict based on actual documented progress rather than vibes.

• second-brain-librarian (Optional): Owns a dedicated, cross-project Obsidian wiki, building and querying reusable Verse mechanics across all your islands.

• second-brain-trainer (Optional): Runs parallelized sweeps using second-brain-scout helpers to rapidly capture reusable patterns from large projects without corrupting vault data.

Key Features & Concrete Developer Benefits

1. Cross-Project Shared Knowledge (uefn-lessons)

• Standard AI project memory resets on new islands.

• This kit includes a shared, system-level skill called uefn-lessons that persists across all projects.

• Agents log Verse, UEFN, and MCP gotchas to a single shared knowledge base. The more projects you build, the smarter your agents get, eliminating blind spots on future islands.

2. Automated Post-Playtest QA & Documentation

• Uses Claude Code hooks tied to UEFN’s MCP call payloads.

• Ending a playtest session automatically triggers qa-regression and planner-docs in the background to log new bugs and update project status without manual effort.

3. Discover-Driven Retention Optimization

• Integrates the discover-retention skill grounded in Epic’s official Discover surfacing metrics (bounce rate, QPTR, and average playtime thresholds).

• Early-game retention recommendations target actionable mechanics that actually drive island visibility on Fortnite’s Discover tab.

4. Real-Time Arcade Agent Console

• Includes a local, browser-based arcade dashboard that lights up active agent cards.

• Tracks live execution times, subagent handoffs, token consumption, and session stats in real time.

5. Token-Aware Progressive Disclosure Skills

• Ships with 7 specialized skills (verse-patterns, uefn-device-gotchas, performance-uefn-checklist, discover-retention, second-brain-query, token-aware-coding, brand-collections-uefn).

• Dense reference guides load only when needed, reducing overall context bloat and token expenditure.

6. Complete Data Transparency & Security

• Project templates sit inside your project’s native Content/ subfolder, fully protected by UEFN’s cloud sync.

• Manual setup steps are intentionally unscripted so you maintain complete transparency over what is installed on your system.

• 

Getting Started

Setup is straightforward and divided into two clear phases:

1. Phase 1 (Environment Setup): Performed once per machine to configure the UEFN MCP server and copy user-level agents and skills to your global profile.

2. Phase 2 (Per-Project Setup): Repeated for each island by copying the template into the project’s Content/ folder.

Get the Free Kit & Join the Community

The entire setup guide and kit are 100% free.

Join the Discord server to download the files and connect:

https://discord.gg/WXF9UgzvRX

If this kit saves you time, support creator code ROOT in the Fortnite Item Shop, and follow along across social channels:

• YouTube: https://www.youtube.com/@Mimmo_the_root

• Twitch: mimmo_the_root - Twitch

• X: mimmo the Root | UEFN | Fortnite | SAC: ROOT (@mimmo_the_root) / X

• TikTok: TikTok - Make Your Day ot

• Instagram: https://www.instagram.com/mimmo_the_r oot

• Epic Games Community Page: https://communities.epicgames.com/community/U1gO/mimmo-the- root

— Mimmo_the_root

[mimmo_the_root]

Dream Bot Team, update — from 5 agents to 12, and now open source

Follow-up to my earlier post about this Claude Code setup for UEFN (that was v1.49). A lot has changed since then, so here’s what’s new — and the big one: it’s now free and open source.

The team grew from 5 to 12 agents. The original five (project-bootstrap, coder, qa-regression, planner-docs, release-gate) are all still there, but three things got added around them:

• A real compliance/review system. Originally there was no independent check that a task actually matched what was asked before calling it done. Now there are three separate gates, each with one job: intent-gate checks before any code is written whether the task is actually clear enough to implement (catches ambiguity before it gets guessed at); intent-reviewer checks the finished work against the spec; compliance-reviewer checks it against this kit’s own mechanical rules (logging, naming, DemoDisplay, deprecated APIs, and so on) — and only runs after the first one passes. No single agent grades its own work anymore, on either axis.
• A plan-first workflow. Every code change now ties to a tracked task in ROADMAP.md before coder touches anything, and a task only gets marked done after both review gates above have signed off — not on coder’s own say-so.
• growth-manager and eight marketing/growth skills — analytics, competitor analysis, launch planning, retention/game-design feedback, trailers, thumbnails, titles/descriptions, update-writing. The kit isn’t just about writing Verse anymore, it also helps with the business side of shipping a map. (still under test, more improvement in future update)
• codebase-auditor — an independent, on-demand whole-codebase quality audit, separate from the per-task review gates above.

Two things I built specifically to make it feel less like a black box:

• A real-time Agent Console — a local dashboard you open in your browser that shows which agent is working right now, a live “miniflow” rail tracking a task through its review-gate sequence, MCP server activity, and rough token usage for the session.
• uefn-lessons and mcp-tool-contracts — the shared, cross-project lessons file from the original post is still the core of it, plus a newer skill that’s the single source of truth for “which exact MCP tool do I use for X,” so the same instruction can’t quietly diverge across different agent files anymore.

And the actual news: it’s open source now. MIT licensed, on GitHub, free to fork, modify, or send a PR against: GitHub - mimmo-the-root/UEFN-dream-bot-team: "Claude Code multi-agent starter kit for UEFN — twelve specialized subagents (each with one narrow job), a plan-first workflow with two independent review gates, a shared cross-project lessons skill, and an optional real-time Agent Console. Free, documented, works for one island or fifty." · GitHub

Full changelog and setup guide are in the repo if you want the blow-by-blow of everything between v1.49 and now. Questions or bugs — open an issue, happy to help.

[mimmo_the_root]

Watching a task move through review, live — the Agent Console’s “miniflow”

Small addition to Dream Bot Team worth its own callout: the Agent Console now has a live rail showing exactly where a task sits in the review pipeline — intent-gate → coder → intent-reviewer → compliance-reviewer → planner-docs — updating in real time as each stage finishes.

Each ring is a stage: green means done, amber and pulsing means it’s the one actually working right now (with a running timer), grey means not reached yet. If a reviewer sends work back, it shows up right there as a small reject counter on that stage instead of just vanishing into the log.

The point isn’t decoration — before this, you had to infer pipeline state from which agent card happened to be lit up. Now it’s just… visible.

Part of the free, open source kit: GitHub - mimmo-the-root/UEFN-dream-bot-team: "Claude Code multi-agent starter kit for UEFN — twelve specialized subagents (each with one narrow job), a plan-first workflow with two independent review gates, a shared cross-project lessons skill, and an optional real-time Agent Console. Free, documented, works for one island or fifty." · GitHub