UEFN-VERSE-MCP
by qfoldit
README.md
# UEFN TOOLBELT
**355 Professional Tools for UEFN Python Integration.**
> Built by **Ocean Bennett** — 2026
[](https://github.com/undergroundrap/UEFN-TOOLBELT/actions/workflows/ci.yml)
[](LICENSE)
[](docs/CHANGELOG.md)
[](https://github.com/undergroundrap/UEFN-TOOLBELT/discussions)
---
> **This is the qFoldIT-extended build — "Universal Builder & Gamifier of Scientific
> [Experiments] with Quantum Computing."**
>
> On top of the 355 stock UEFN Python tools documented below, this repository also
> ships, all under [`qfoldit/`](qfoldit/README.md) and [`.claude/skills/`](.claude/skills/):
>
> - a **default-deny IP compliance gate** in front of `run_toolbelt_tool`/`execute_python` —
> entertainment brands (LEGO, TMNT, Star Wars, ...) plus, opt-in-by-default in the live
> server, real-world **scientific-equipment** (Thermo Fisher, Agilent, Bruker, ...) and
> **vehicle** (Sherp, Kamaz, Caterpillar, ...) trademark watchlists;
> - two real **protein-folding science pipelines** — a classical quantum-walk-inspired
> Metropolis simulation, and an import-guarded real CVaR-VQE wrapper that reports
> `unavailable` rather than fabricating a result when its optional dependency isn't installed;
> - a deterministic, sha256-seeded **gamification layer** turning a science result into
> levels/par-scores/achievements, with no LLM call and no invented parameters;
> - an engine-neutral **scene-graph pipeline** (UAG) plus a UEFN-specific builder skill that
> only calls real, grep-verified toolbelt tools and reports — rather than papers over — the
> node types it can't yet realize (camera actors, UI panels);
> - a **reproducible, licensing-cleared, publication-honest Experiment Record system** that
> ties all of the above together: every game object placed while building a scene has its
> licensing decision automatically collected, every science result carries a
> citation-grounded methods section that never mislabels classical results as quantum, and
> every record is durably logged, not just returned once and lost.
>
> None of this is stock UEFN Toolbelt. See
> **[qFoldIT Trust & Compliance Layer](#qfoldit-trust--compliance-layer)** and
> **[qFoldIT Science, Gamification & Experiment Pipeline](#qfoldit-science-gamification--experiment-pipeline)**
> below for the full picture, or **[`qfoldit/README.md`](qfoldit/README.md)** for the
> folder-level detail and honest list of what's still a documented gap rather than done.
---
> **Historic Discovery — March 22, 2026**
>
> On this date, **Ocean Bennett** became the first person to programmatically catalogue
> the complete Fortnite Creative device palette using UEFN Python's Asset Registry —
> **4,698 Creative device Blueprints across 35 categories**, extracted from 24,926 total
> Blueprint assets in the sandboxed UEFN environment.
>
> Prior to this, no public tool, script, or documentation existed that mapped the full
> set of placeable Creative devices accessible from Python. Epic's own documentation
> covers individual devices. This is the first machine-readable index of the entire palette.
>
> The scan also uncovered a critical silent failure pattern in UEFN 40.00's Asset Registry
> API: deprecated `AssetData` properties (`object_path`, `asset_class`) silently throw
> inside a `try/except` block, causing every asset to be skipped with no error message —
> a bug that would defeat any developer who didn't know to split the exception handling.
>
> The full technical breakdown, the category table, and the discovery story are in
> **[docs/AI_AUTONOMY.md](docs/AI_AUTONOMY.md)**.
---
Automate the tedious, script the impossible, and bridge the gap between Python and Verse.
UEFN Toolbelt is a master utility designed to leverage the **2026 UEFN Python 3.11 Update**,
allowing creators to manipulate actors, manage assets, and generate boilerplate Verse code
through a high-level, developer-friendly interface — all from a single persistent menu entry
in the UEFN editor bar. **355 registered tools** across 55 categories, complete AI-agent
readiness (100% structured dict returns), and a unified theme system so every window in the
platform looks and feels identical.
---
## The Demo — New Project Setup in Under a Minute
> Proven live. Scaffold + Verse game manager + 493-actor arena + reusable room stamps. No manual steps except one Build click.
```python
# Command 1 — professional folder structure + Verse game manager deployed
tb.run("project_setup", project_name="MyGame")
# → 56 folders created in Content Browser
# → MyGameManager.verse generated and deployed to project
# Command 2 — symmetrical Red vs Blue arena spawns in the viewport
tb.run("arena_generate", size="medium")
# → 493 actors placed (6 Red spawns, 6 Blue spawns)
# Command 3 — save a loot room you built as a reusable stamp
tb.run("stamp_save", name="loot_room")
# → saved 12 actors to Saved/UEFN_Toolbelt/stamps/loot_room.json
# Navigate camera to a new location, then stamp it in
tb.run("stamp_place", name="loot_room", yaw_offset=90)
# → 12 FortStaticMeshActors placed at camera, rotated 90°
# One click — Verse menu → Build Verse Code
# Command 4 — error loop closes automatically
tb.run("verse_patch_errors")
# → build_status: SUCCESS
```
What used to take 30+ minutes of manual setup runs in seconds. Scaffold, code, layout,
stamp your best rooms, build — ready to publish.
---
## 🤖 Claude Can Build Your Game — Autonomously
> **This is the headline feature.** Not "AI helps you code" — AI *reads your live level, writes the Verse, and deploys it*. Zero copy-paste. Zero manual wiring.
Here is the complete autonomy loop running live on a real project with 521 actors:
**Step 1a — Claude reads the entire level:**
```python
tb.run("world_state_export")
# → Captured 521 actors. Saved to docs/world_state.json
```
**Step 1b — Claude reads every device available in Fortnite (not just what's placed):**
```python
tb.run("device_catalog_scan")
# → Scanned 24,926 Blueprint assets across all Fortnite packages
# → 4,698 Creative devices identified across 35 categories
# → (Timer ×8, Capture ×7, Score ×28, Spawner ×94, Camera ×446, NPC ×173, ...)
# → docs/device_catalog.json — Claude's complete device palette, git-tracked
```
**Step 2 — Claude identifies all Creative devices and generates a wired game skeleton:**
Claude reads `world_state.json`, finds every device in the level (`FortCreativeTimerDevice`,
`capture_area_device`, `guard_spawner_device`, `button_device`, 14 teleporters, etc.),
and writes a complete `creative_device` class — `@editable` declarations, `OnBegin` event
wiring, round flow, creature waves, capture logic, and clean shutdown — all referencing the
**actual device labels from your level**.
**Step 3 — Claude deploys it directly to the project:**
```python
tb.run("verse_write_file", filename="device_api_game_manager.verse", content=verse_code)
# → Written: Device_API_Mapping\Verse\device_api_game_manager.verse (6187 bytes)
```
**Step 4 — Build Verse. First try.**
```
VerseBuild: SUCCESS -- Build complete.
```
No type errors. No manual editing. A fully wired, compilable Verse game manager — generated
from live level state in under 60 seconds.
**What the generated skeleton contains:**
- `@editable` refs for every device discovered: Timer, Timed Objective, Round Settings, Capture Area, 2× Item Spawners, Creature Spawner, Creature Manager, Creature Placer, 2× Buttons, 2× Conditional Buttons, Lock Device, 5 keyed Teleporters, Audio Mixer, Weapon Mod Bench
- `OnBegin` that starts the timer, spawns async watchers, and subscribes to all button/timer events
- `WatchCaptureArea()` — async loop: team captures zone → spawn reward items → trigger creature wave
- `WatchCreatureWaves()` — async loop: enable creature manager → 30s cooldown → repeat
- `OnButtonPressed` / `OnButton2Pressed` — unlock doors, activate conditional gates
- `EndRound()` — stops timer, disables creature manager, despawns guards, stops audio
Place the device in your level, drag your real actors into the `@editable` slots, and play.
> See [docs/PIPELINE.md](docs/PIPELINE.md) for the complete 6-phase industrial pipeline —
> every tool, every phase, Claude's full execution script, and the recursive build+fix loop.
> See [docs/AI_AUTONOMY.md](docs/AI_AUTONOMY.md) for the live run technical breakdown.
---
## 🤖 AI-Accelerated Development (One-Click Sync + Tool Manifest)
Toolbelt is built to be used with AI (Claude/Gemini). To give your AI **perfect information** about your project's unique Verse devices and custom props:
1. **Open Dashboard**: Run `tb.launch_qt()` or use the `Toolbelt` menu.
2. **One-Click Sync**: Click the **"Sync Level Schema to AI"** button in **Quick Actions**.
3. **Instant Content**: The 1.6MB schema is automatically copied to your `docs/` folder. Your AI now knows every hidden property in your specific level.
4. **Export Tool Manifest**: Run `tb.run("plugin_export_manifest")` to write `Saved/UEFN_Toolbelt/tool_manifest.json` — a machine-readable index of all 358 tools with their parameter signatures, types, defaults, and categories. An AI agent can load this file and immediately know what every tool does and how to call it, without reading source code.
---
## Table of Contents
- [Claude Can Build Your Game — Autonomously](#-claude-can-build-your-game--autonomously)
- [Concept](#concept)
- [The Workflow Loop — Efficiency Mechanic](#the-workflow-loop--efficiency-mechanic)
- [Tech Stack](#tech-stack)
- [Architecture Overview](#architecture-overview)
- [Directory Structure](#directory-structure)
- [Key Systems — How They Work](#key-systems--how-they-work)
- [Math & Performance Scaling](#math--performance-scaling)
- [Tool Reference](#tool-reference)
- [Getting Started](#getting-started)
- [Adding a New Tool](#adding-a-new-tool)
- [Custom Plugins & Security](#custom-plugins--security)
- [qFoldIT Trust & Compliance Layer](#qfoldit-trust--compliance-layer)
- [qFoldIT Science, Gamification & Experiment Pipeline](#qfoldit-science-gamification--experiment-pipeline)
- [API Capability Crawler](#api-capability-crawler)
- [Fortnite Device API Mapping](#fortnite-device-api-mapping)
- [MCP — Connect Any AI to UEFN](#mcp--connect-any-ai-to-uefn)
- [Spec-Accurate Verse Code Generation](#spec-accurate-verse-code-generation)
- [CLAUDE.md — Instant AI Context](#claudemd--instant-ai-context)
- [Why This Is the Best UEFN Python Tool](#why-this-is-the-best-uefn-python-tool)
- [Documentation](#documentation)
- [Quick-Reference Command Table](#quick-reference-command-table)
- [Patch Notes](#patch-notes)
- [Contributing](#contributing)
- [Tool Requests](#tool-requests)
- [License](#license)
---
## Security
> **Why this matters:** The UEFN Python community has already seen proof-of-concept malicious
> scripts that flood the editor with spam windows when run from untrusted sources.
> Python editor scripts run with full access to your project — treat them like executable code.
### This repo's safety guarantees
| Practice | How it's enforced here |
|---|---|
| **Audited code** | Every line in this repo is open-source and reviewable |
| **No network calls** | Zero outbound HTTP/socket connections anywhere in the codebase |
| **No file writes outside project** | All output goes to `Saved/UEFN_Toolbelt/` inside your project |
| **No eval/exec on external input** | No dynamic code execution from files, URLs, or user strings |
| **Undo-safe** | Every destructive operation is wrapped in `ScopedEditorTransaction` |
### How to verify before running any `.py` file (from this repo or anyone's)
```python
# Before exec()-ing any script, read it first:
with open("path/to/script.py") as f:
print(f.read())
# Red flags to look for:
# - import requests / import http / import socket
# - exec() / eval() on anything fetched from outside
# - os.system() / subprocess with network targets
# - anything writing outside your project folder
```
### Running untrusted community scripts safely
1. Always read the script before running it
2. Run unknown scripts in a **throwaway project** first
3. Keep UEFN's Python access set to **Editor Scripting only** (the default)
---
## qFoldIT Trust & Compliance Layer
> Merged from the [qFoldIT](https://github.com/qfoldit) team — a default-deny gate that sits in
> front of `run_toolbelt_tool` and `execute_python` in `mcp_server.py`, so that generation/import
> calls involving third-party brand IP (LEGO, TMNT, Star Wars, The Walking Dead Universe, Squid
> Game, KPop Demon Hunters, real-world scientific-equipment brands, real-world vehicle brands, and
> others) can't slip past this repo's own compliance requirements — for authorized creators, and
> enforced the same way against any MCP server that connects.
**Everything qFoldIT-related lives under [`qfoldit/`](qfoldit/README.md)** — compliance, science,
and monetization each in their own subfolder, data files next to the code that reads them, runtime
logs under `qfoldit/logs/`. Two Claude Skills also live under
[`.claude/skills/`](.claude/skills/game-designer) (`game-designer`, `unreal-world-builder`). See
[`qfoldit/README.md`](qfoldit/README.md) for the full folder map; tests stay in the top-level
`tests/` folder alongside `smoke_test.py` (85 qFoldIT tests total, plain `python3`).
### What it actually does
| Layer | Question it answers | How |
|---|---|---|
| **IP Trust Gate** | "Is this asset call actually licensed?" | Default-deny watchlist match → requires a real, dated `qfoldit/compliance/license_manifest.json` entry, not just presence in the project. The live server runs `TrustRuntime.with_extended_watchlist()`, so the watchlist covers entertainment IP **plus** real-world scientific-equipment and vehicle trademarks (opt-in extensions, active by default in `mcp_server.py`) |
| **Provenance verification** | "Did this asset really come from where the call claims?" | When an `asset_metadata_fn` is wired to the live editor (via the existing `get_asset_info` IPC command — no network call), path claims are checked against the engine's own Asset Registry, not trusted as text. A path that *looks* Epic-native but resolves somewhere else is blocked as a spoofing attempt, not silently allowed |
| **Science MCP Registry** | "Is this connected MCP server safe/stable enough to auto-run?" | `verified` / `connected` auto-connect; `best_effort` requires an explicit opt-in; `reference_only` never auto-connects |
| **Science pipeline** | "Can I get a real (or honestly-unavailable) folding result from inside the editor?" | `qfoldit_quantum_walk_fold` (real classical Metropolis simulation) and `qfoldit_quantum_vqe_fold` (real CVaR-VQE, import-guarded — reports `unavailable` rather than fabricating an energy) — see [qFoldIT Science, Gamification & Experiment Pipeline](#qfoldit-science-gamification--experiment-pipeline) below |
| **Scene-object licensing collection** | "Was every object placed while building a scene actually checked, not just the science calls?" | `qfoldit_scene_build_start()` marks the start of a build; `TrustRuntime.decisions_since()` reads the runtime's own audit log back out; `qfoldit_build_experiment_record(auto_collect_licensing_since=...)` folds every real decision made during that window into the record automatically |
| **Monetization Gate** | "Can we accept payment for this commission, and what will it cost us?" | Off-platform ("Cameo-style") commissions run through the same IP watchlist before payment; Boltz-backed requests get a **local, offline** compute-time estimate from `qfoldit/monetization/boltz_pricing.json` — no dollar figure is ever fabricated: `rate_usd_per_gpu_second` ships as `null` until you fill it in from your real Modal billing dashboard |
| **Dashboard visibility** | "What has the gate actually blocked/allowed/priced?" | `qfoldit_trust_dashboard` tool (category: Dashboard) — read-only panel summarizing the audit log, licensed brands, and commission ledger; `qfoldit_list_experiment_records` for the durable experiment-record log specifically |
### Why provenance checking matters here
A text match alone (`"lego" in payload`) can't tell a genuine LEGO-template asset from a custom
import that merely mentions the word "lego" in a variable name. `TrustRuntime.verify_asset_provenance()`
closes that gap when wired to the live engine: every path-like reference in a call's arguments is
resolved through the real Asset Registry, and the **engine's answer always overrides the text's
claim**. Manifest entries additionally support `content_plugin_ids` — the real plugin/content-pack
identifier(s) the engine reports for a brand's genuine template assets — for an even tighter check
than namespace prefixes alone. These start empty (`[]`) per brand; fill them in from a real Content
Browser inspection, they are never guessed.
### Real-world scientific-equipment and vehicle trademarks
Beyond entertainment IP, `qfoldit/compliance/watchlists/scientific_equipment_watchlist.json` and
`vehicle_watchlist.json` cover lab/analytical-equipment brands (Thermo Fisher, Agilent, Danaher's
sub-brands, Bruker, Illumina, ...) and vehicle/heavy-equipment brands (Sherp, Kamaz, Caterpillar,
Scania, Tesla, ...). Neither category has an Epic IP Partner Licensing Agreement pathway, so a match
resolves to **blocked** unless a real, independently-negotiated `license_type='creator_independent_license'`
manifest entry exists — no entry is ever added speculatively. Several generic-word brand names (e.g.
a manufacturer whose name transliterates to the Russian word for "north") are documented but shipped
`enabled: false` to avoid false-positive noise from ordinary words. Every term/alias in these files is
English/Latin-script by design — see the files' own `_README` block for the resulting Cyrillic-text
coverage trade-off. Full detail: [`qfoldit/compliance/watchlists/`](qfoldit/compliance/watchlists/).
### Boltz commission pricing — honest by construction
This repo's own [Security](#security) guarantees zero outbound network calls anywhere in the
codebase, so commission pricing can't be a live API lookup either. `estimate_boltz_cost()` computes
a compute-time estimate locally from sequence length and a rate table in `boltz_pricing.json`. Until
`rate_usd_per_gpu_second` is filled in from your actual Modal billing, `qfoldit_evaluate_commission`
returns `estimated_cost_usd: null` — never a made-up number, and never silently treated as free.
## qFoldIT Science, Gamification & Experiment Pipeline
### Science
Two real, tested science tools, callable from inside the editor via `mcp_server.py`, both gated
through the same IP compliance check as everything else here:
- **`qfoldit_quantum_walk_fold(sequence, steps, continuous_space, seed)`** — a genuine classical
Metropolis simulation (biased random-walk proposal + real acceptance step) over peptide torsion
angles, inspired by the QFold algorithm (Casares et al., *Quantum Sci. Technol.* 7, 025013, 2022)
but explicitly **not** quantum — no circuit simulator or hardware involved. Reconstructs a real
3D backbone via standard NeRF geometry. **Documented, test-confirmed limitation:** single-
dihedral moves can leave chains of ~14+ residues kinetically trapped in a self-clashed
conformation, so a longer run isn't guaranteed to reach lower energy than a short one at that
length — see the function's docstring and `tests/test_qfoldit_quantum_runner.py`'s
`test_known_limitation_longer_chains_can_stay_kinetically_trapped` for the confirmed numbers.
- **`qfoldit_quantum_vqe_fold(sequence, alpha, shots)`** — a wrapper around the real CVaR-optimized
VQE from QuPepFold (Uttarkar et al., *PLOS ONE*, 2026). Requires the separate `qupepfold` package
(Qiskit/Braket-dependent); if it's not installed, returns `status: "unavailable"` with an
`install_hint` — never a fabricated ground-state energy. Not independently verified against a real
installed `qupepfold` release at authoring time — this caveat is carried forward automatically into
any experiment record built from this tool's output (see below).
### Gamification and the scene pipeline (UAG)
- **`qfoldit_generate_game_design(source_json, title, difficulty)`** — deterministic (sha256-seeded,
no LLM call) generator turning either science tool's output into levels/par-scores/achievements, so
any in-game score traces back to the exact number in the source science result.
- **`qfoldit_gamedesign_to_uag_seed(game_design_json)`** — bridges a game design document into a
starter [Universal Assembly Graph](.claude/skills/game-designer/references/uag_schema.md) (UAG):
one `group` node per level, no meshes/lights/interactions invented. Hand the result to the
**`game-designer`** Claude Skill ([`.claude/skills/game-designer/`](.claude/skills/game-designer))
to design the actual scene content.
- **`unreal-world-builder`** ([`.claude/skills/unreal-world-builder/`](.claude/skills/unreal-world-builder))
— the UEFN-specific engine adapter that realizes a validated UAG by calling this repo's own
`run_toolbelt_tool` with real tool names (`light_place`, `zone_spawn`, `audio_place`,
`niagara_spawn_system`, `import_fbx`, `stamp_place`, `device_set_property`, ...). It documents
rather than papers over its gaps — no tool currently spawns a persistent `CameraActor` or a Verse
UI panel, so a UAG needing those gets an explicit report, not a silent skip.
### Experiment Record — reproducibility + licensing + publication readiness, tied together
`qfoldit_build_experiment_record` is the tool that operationalizes this build's stated goal: turning
gameplay into a reproducible scientific experiment where every game object, action, and computation
corresponds to a licensing decision, a scientific-validity statement, and a publication-readiness
check.
- **Reproducibility** — `experiment_id` is a sha256 hash over exactly the deterministic
inputs/outputs (the science result + its reproduction parameters + any game-design seed); two
records with the same id are, by construction, the same experiment.
- **Licensing, for every object in the scene, not just science calls** —
`qfoldit_scene_build_start()` marks the start of a build; `unreal-world-builder` places the scene;
`qfoldit_build_experiment_record(auto_collect_licensing_since=...)` then reads
`TrustRuntime`'s own audit log back out (`decisions_since()`) and folds every real decision made
during that window into the record automatically. A blocked decision anywhere in that window flags
`publication_checklist.licensing_all_cleared: false`.
- **Publication readiness, reported not decided** — `methods_section()` pulls from a small,
hand-checked citation table keyed to the *actual* algorithm, not the tool's marketing name: the
classical Metropolis "quantum_walk" simulation is always described as classical (never as a quantum
computation), and the real VQE path always carries its "not independently verified" caveat forward.
`publication_checklist()` returns five named boolean checks — never a single yes/no verdict; a human
still makes the actual publication call.
- **Durable** — every record built is appended to `qfoldit/logs/experiment_records.log.jsonl`
(`qfoldit_list_experiment_records` reads it back), so it survives past the one response that
returned it.
```bash
python3 tests/test_qfoldit_trust_runtime.py # 16/16 — includes engine-provenance cases
python3 tests/test_qfoldit_watchlist_extensions.py # 8/8 — sci-equipment/vehicle watchlists
python3 tests/test_qfoldit_science_mcp_registry.py # 7/7
python3 tests/test_qfoldit_monetization_registry.py # 8/8 — includes Boltz pricing cases
python3 tests/test_qfoldit_quantum_runner.py # 12/12 — includes the kinetic-trapping case
python3 tests/test_qfoldit_gamedesign.py # 8/8 — includes a real end-to-end run
python3 tests/test_qfoldit_uag_bridge.py # 9/9 — includes a validator sync-check
python3 tests/test_qfoldit_experiment_record.py # 9/9 — includes a real end-to-end run
python3 tests/test_qfoldit_scene_licensing_collection.py # 8/8 — audit-log read-back + persistence
# 85/85 qFoldIT tests total
```
---
## ⚠️ Automated Integration Testing
The smoke test verifies that all tools *register* correctly. The integration test verifies that they *work* — in a live UEFN editor, against real actors.
### What the integration test actually does
`toolbelt_integration_test` is a fixture-driven test harness that runs inside UEFN:
1. **Spawns** temporary cube/sphere actors programmatically
2. **Selects** them via `EditorActorSubsystem`
3. **Runs** each tool against that live selection
4. **Verifies** the result — property changed, actor spawned, file written, count correct
5. **Cleans up** every actor it touched via a single undo transaction
**163 tests across 358 tools** — covering materials, bulk ops, patterns, scatter, splines, snapshots, asset management, Verse tools, screenshots, LODs, arena, measurement, localization, zones, stamps, actor org, proximity placement, advanced alignment, signs, post-process, audio, level health, config, lighting, and world state.
This is the closest thing to a full CI suite possible inside the UEFN Python sandbox. If this passes, you have high confidence that the core tool logic is sound — not just that it imported.
### How to run it
```python
# Use a clean template level — never run in a production project
import UEFN_Toolbelt as tb; tb.register_all_tools(); tb.run("toolbelt_integration_test")
```
Results are written to:
```
Saved/UEFN_Toolbelt/integration_test_results.txt
```
If the editor crashes mid-run (rare), the file will contain partial results up to the last completed test — useful for pinpointing which section caused the crash.
> [!WARNING]
> **DO NOT run the full integration test in a live production project.**
> It spawns, modifies, and deletes actors. Use a blank **Empty Level** or throwaway test project.
> Every test section cleans up after itself, but the undo history will be long.
### Smoke test vs integration test
| | Smoke Test | Integration Test |
|---|---|---|
| Runs outside UEFN? | No (needs editor) | No (needs editor) |
| Tests all 358 tools? | Registry only | Live execution |
| Requires level actors? | No | Yes (spawns its own) |
| Safe in production? | Yes | **No — use blank level** |
| Runtime | ~5 seconds | ~35 seconds |
| Command | `tb.run("toolbelt_smoke_test")` | `tb.run("toolbelt_integration_test")` |
Run the smoke test after every change. Run the integration test before submitting a PR.
---
> **Community context:** Built in response to the March 2026 UEFN Python wave. Inspired by
> early community tools like standalone material editors, spline prop placers,
> and Verse device editors — then taken further into a unified, permanently
> docked creator toolkit.
---
## Concept
UEFN Toolbelt isn't just a library — it's a **meta-framework**. While the standard UEFN
Python API provides the *what*, Toolbelt provides the *how*. It follows a
**Blueprint-to-Script philosophy**: taking complex manual editor tasks (mass material
assignment, procedural arena generation, bulk Verse device editing) and reducing them to
single-line Python commands callable from a top-menu dropdown.
The toolbelt is designed for three creator personas:
| Persona | Problem Solved |
|---|---|
| **The Automator** | Needs to spawn and configure hundreds of actors across a large map in seconds |
| **The Integrator** | Building bridges between Python data and Verse-driven gameplay systems |
| **The Janitor** | Mass-cleaning naming violations, missing materials, misaligned props |
---
## 🏗️ Solving Soul-Crushing Repetition (The 8 Pillars)
The UEFN Toolbelt exists because manual work doesn't scale, but scripts do. We solve 90% of the repetitive tasks reported by the UEFN community:
1. **Manual Placement & Spacing**: Procedural spawners, align/distribute scripts, and spline-based distribution.
2. **Asset Organization**: Bulk import pipelines, smart renaming, and auto-folder scaffolding.
3. **Material Batch Edits**: Parameter randomization, team-color splitters, and bulk texture swaps.
4. **Structural Elements**: High-performance cable, wire, fence, and rail generators.
5. **Layout Math**: Instant 3D distance and travel-time (Walk/Run/Sprint) estimation.
6. **Bulk Optimization**: Automated LOD generation, memory audits, and cooking pre-flight checks.
7. **Verse Boilerplate**: Selection-to-code generation and spec-accurate device stubs.
8. **Component Spam**: Scripted "Quick-Add" macros for entity sets and Niagara effects.
9. **AI-Agent Readiness**: Every tool returns a structured `dict` — AI agents operate on results, not logs. The MCP bridge, `tool_manifest.json`, and `describe_tool` command let any LLM discover and call every tool autonomously. This is the first UEFN toolkit designed to be equally usable by humans and AI agents.
---
## AI Return Contract — 100% Machine-Readable
Every single one of the 358 tools returns a clean, structured `dict`. No `None` returns. No bare primitives. No output that requires parsing log lines. This was completed as Phase 21 and applies to every tool added since.
### What Claude gets back
```python
# Asset scan
{"status": "ok", "count": 4, "textures": [
{"name": "T_Rock", "path": "/Device_API_Mapping/Textures/T_Rock",
"compression": "TC_DEFAULT", "srgb": "true", "size_x": "2048", "size_y": "2048"}
]}
# Audit
{"status": "ok", "total": 12, "clean": 10, "issues": 2, "issue_list": [
{"name": "SK_Hero", "path": "/Device_API_Mapping/Characters/SK_Hero",
"issues": ["No physics asset assigned"]}
]}
# Write op (dry_run)
{"status": "ok", "dry_run": true, "changed": 8, "skipped": 3, "changes": [
{"name": "T_Wall", "from": "TC_DEFAULT", "to": "TC_NORMALMAP"}
]}
# Error — always catchable
{"status": "error", "message": "Could not load asset at '/Game/Missing'."}
```
### The full MCP chain
```
Claude Code → MCP server → run_tool → registry.execute() → _serialize(result) → JSON response → Claude reads it
```
Claude never has to parse a log line. Every domain follows the same shape:
| Domain | Keys Claude reads |
|---|---|
| Asset scans | `count` + array with `name/path/type` per asset |
| Audits | `total/clean/issues` + `issue_list` with per-item details |
| Write ops | `dry_run` + `changed/skipped` + `changes` array |
| Level actors | `count` + array with `label/class/location` |
| Inspections | Full property dict for the specific asset |
| Exports | `output_path` + `count` |
| Errors | `{"status": "error", "message": "..."}` — always catchable |
### Tool discovery before first call
`tb.run("plugin_export_manifest")` writes `Saved/UEFN_Toolbelt/tool_manifest.json` — a machine-readable index of all 358 tools with full parameter signatures (name, type, required, default) and a concrete `example` call string for every tool. Claude reads this once and knows exactly what params to pass to any tool without guessing.
---
## The Workflow Loop — Efficiency Mechanic
This is the core philosophy. We solve the **"Iteration Tax"** — the time lost between an
idea and its implementation in the editor.
### The Three Efficiency Scalars
$$T_{\text{save}} = (T_{\text{manual}} \times N) - (T_{\text{script}} + T_{\text{execution}})$$
| Scalar | Value | Description |
|---|---|---|
| **Batch Multiplier** | $1.5^N$ | Time saved grows super-linearly as task count N rises |
| **Boilerplate Reduction** | $\approx 0.8 \times \text{Lines}$ | Toolbelt reduces raw UE Python verbosity by ~80% |
| **Net Speed Gain** | ~300% | Measured on world-building tasks vs. manual editor clicks |
---
## Tech Stack
| Layer | Technology | Purpose |
|---|---|---|
| **Core** | Python 3.11 | Native language of the UEFN experimental integration |
| **API Wrapper** | `unreal` module | Direct hooks into `EditorActorSubsystem`, `EditorAssetLibrary`, `MaterialEditingLibrary` |
| **Undo Safety** | `unreal.ScopedEditorTransaction` | Every destructive operation is wrapped — one Ctrl+Z removes anything |
| **UI** | Editor Utility Widgets (UMG) | Dockable dashboard with tabs; buttons fire `Execute Python Command` nodes |
| **Verse Bridge** | `verse_snippet_generator.py` | Generates `.verse` files from Python context (selected actors, level state) |
| **Data** | JSON | Custom preset storage and import logs in `Saved/UEFN_Toolbelt/` |
| **Menu** | `unreal.ToolMenus` | Permanent top-bar menu entry injected on editor startup |
---
## Architecture Overview
```
UEFN Editor
│
│ init_unreal.py ← auto-executed by UEFN on every startup
│ (generic submodule loader — NOT part of the Toolbelt package)
│ If you already have one, add the package-discovery pattern
│ from the provided template instead of overwriting yours.
▼
UEFN TOOLBELT ─── UEFN_Toolbelt/__init__.py (register, launch, run, registry accessor)
│
├── core.py Shared utilities: undo_transaction, get_selected_actors,
│ with_progress, color_from_hex, spawn_static_mesh_actor, …
│
├── registry.py ToolRegistry singleton — @register_tool decorator,
│ execute-by-name, category listing, tag search,
│ to_manifest() for full parameter-introspected export
│
└── tools/
├── material_master.py 17 presets, gradient, harmony, team split, save/load
├── arena_generator.py Symmetrical Red/Blue arenas (S/M/L)
├── spline_prop_placer.py Props along splines — count or distance mode
├── bulk_operations.py Align, distribute, randomize, snap, mirror, stack
├── verse_device_editor.py List, filter, bulk-edit, export Verse devices
├── smart_importer.py FBX batch import + auto-material + Content Browser organizer
├── verse_snippet_generator.py Context-aware Verse boilerplate from level selection
├── text_painter.py Colored 3D text actors with saved style presets
├── asset_renamer.py Epic naming convention enforcer with dry-run + audit
├── project_scaffold.py Professional folder structure generator (4 templates)
├── verse_schema.py Verse Digest IQ & Universal Schema Search (143th Tool)
├── system_build.py Automated UEFN Build Scraper & Error Monitor
├── system_perf.py Background CPU Optimizer & Monitor
├── measurement_tools.py Distance Calculator & Travel Time Estimator (Phase 15)
└── localization_tools.py Multi-language Text Export/Import (Phase 15)
```
All heavy logic lives in Python. The optional UMG dashboard calls into it via
`Execute Python Command` Blueprint nodes — zero coupling, infinitely extensible.
---
## Directory Structure
```
[YourUEFNProject]/
├── Content/
│ ├── Python/
│ │ ├── init_unreal.py ← COPY HERE if you don't have one yet
│ │ │ (generic loader — not Toolbelt-specific)
│ │ │ ⚠️ If you already have this file, do NOT
│ │ │ overwrite it. Add the package-discovery
│ │ │ loop from the template into your existing file.
│ │ └── UEFN_Toolbelt/ ← COPY HERE — the full package
│ │ ├── __init__.py
│ │ ├── core.py
│ │ ├── registry.py
│ │ └── tools/
│ │ ├── __init__.py
│ │ ├── material_master.py
│ │ ├── arena_generator.py
│ │ ├── spline_prop_placer.py
│ │ ├── bulk_operations.py
│ │ ├── verse_device_editor.py
│ │ ├── asset_importer.py # NEW: URL & Clipboard image importing
│ │ ├── verse_snippet_generator.py
│ │ ├── text_painter.py
│ │ ├── asset_renamer.py
│ │ ├── procedural_geometry.py # NEW: Wire & Volumetric generators
│ │ ├── text_voxelizer.py # NEW: 3D Text geometry generation
│ │ ├── smart_organizer.py # Proprietary Heuristics Engine
│ │ ├── system_perf.py # Background CPU Optimizer
│ │ ├── verse_schema.py # NEW: Verse Digest IQ (Phase 14)
│ │ ├── system_build.py # NEW: Automated Build Monitor (Phase 14)
│ │ ├── measurement_tools.py # NEW: Distance & Travel Time (Phase 15)
│ │ └── localization_tools.py # NEW: Text Export & Translation (Phase 15)
│ └── UEFN_Toolbelt/
│ ├── Blueprints/
│ │ └── WBP_ToolbeltDashboard ← optional EUW — create in UEFN
│ └── Materials/
│ └── M_ToolbeltBase ← required for Material Master
├── deploy.bat ← double-click to deploy to any UEFN project
└── README.md
```
> **Critical path:** Files must live under `Content/Python/` exactly.
> Any other location (e.g. `Content/Scripts/`) is **not** scanned by UEFN's Python interpreter.
---
## Key Systems — How They Work
### 1. The Tool Registry
Every tool self-registers with a one-line decorator. The registry is a singleton shared
across the entire session.
```python
from UEFN_Toolbelt.registry import register_tool
@register_tool(
name="my_tool",
category="Procedural",
description="Does something amazing",
tags=["procedural", "spawn"],
)
def run(**kwargs) -> dict:
...
return {"status": "ok", "count": placed}
```
Calling `tb.run("my_tool")` from anywhere — REPL, Blueprint node, another tool — routes
through the registry with full error containment. A crashing tool never kills your session.
Every tool returns a structured `dict` with at minimum a `"status"` key. This is the
**MCP Return Contract**: AI agents calling tools via the bridge read the return dict directly
from the JSON response. Zero log parsing required.
---
### 2. Material Master
17 built-in presets stored as pure data dictionaries. All material work happens through
`MaterialEditingLibrary` — no Blueprint graph required.
```python
tb.run("material_apply_preset", preset="chrome") # apply to selection
tb.run("material_gradient_painter", # world-space gradient
color_a="#0044FF", color_b="#FF2200", axis="X")
tb.run("material_team_color_split") # auto Red / Blue by X position
tb.run("material_save_preset", preset_name="MyLook") # save to JSON
```
Custom presets persist across sessions in `Saved/UEFN_Toolbelt/custom_presets.json`.
---
### 3. Arena Generator
Instant symmetrical competitive arenas. What used to take 45 minutes of manual prop
placement now runs in under 5 seconds.
```python
tb.run("arena_generate", size="large", apply_team_colors=True)
```
Internally places floor tiles, perimeter walls, a center platform, and team spawn pads —
all in a single `ScopedEditorTransaction`. One Ctrl+Z removes the entire arena.
---
### 4. The Verse Bridge
The biggest pain point in UEFN is syncing editor state with Verse code. The snippet
generator reads your **actual level selection** and produces strongly-typed output:
```python
# Select 6 devices in the viewport, then:
tb.run("verse_gen_device_declarations")
```
Output (written to `Saved/UEFN_Toolbelt/snippets/` and copied to clipboard):
```verse
# Actor: 'TriggerDevice_01' @ (1200, 400, 0)
@editable triggerdevice_01 : trigger_device = trigger_device{}
# Actor: 'SpawnPad_Red_01' @ (3200, 0, 0)
@editable spawnpad_red_01 : player_spawner_device = player_spawner_device{}
```
---
### 5. Undo Safety
Every destructive operation is wrapped in `core.undo_transaction()`:
```python
# From core.py — used by every tool:
with undo_transaction("Material Master: Apply chrome"):
for actor in actors:
_apply_preset_to_actor(actor, preset_data, "chrome")
```
If something goes wrong mid-operation the transaction closes cleanly. No corrupted
undo stack, no phantom changes.
---
### 6. Project Scaffold
The first tool you run on any new UEFN project. Four opinionated templates — pick
the one that matches your workflow:
```python
# See what it would create before touching anything
tb.run("scaffold_preview", template="uefn_standard", project_name="MyIsland")
# One command builds 50+ folders in the right hierarchy
tb.run("scaffold_generate", template="uefn_standard", project_name="MyIsland")
```
Templates are stored as plain JSON in `Saved/UEFN_Toolbelt/scaffold_templates.json`
— email or Git the file to every teammate so everyone gets the same structure instantly.
Custom templates are one call:
```python
tb.run("scaffold_save_template",
template_name="StudioDefault",
folders=["Maps/Main", "Materials/Master", "Materials/Instances",
"Meshes/Props", "Verse/Modules", "Audio/SFX"])
```
---
### 7. Smart Importer
Drop a folder of FBXs and get back a fully organized Content Browser:
```python
tb.run("import_fbx_folder",
folder_path="C:/MyAssets/Props/",
apply_material=True,
place_in_level=False)
```
Assets land in `/Game/Imported/[date]/Meshes/` with auto-generated material instances.
An import log is appended to `Saved/UEFN_Toolbelt/import_log.json` after every run.
---
### 8. Advanced Project Intelligence (Phase 14)
The Toolbelt now features a native "Off-Engine IQ" that understands your Verse code without needing a live actor in the level.
- **Verse Schema IQ**: Parses your `.digest.verse` files to build a mapping of all Verse classes, properties, and events.
- **Build Monitor**: Triggers a background UEFN build and scrapes the log for Verse compilation errors with file/line precision.
- **Global Safety Gate**: A centralized protection layer (`core.safety_gate`) that all "Write" operations must pass through. It prevents accidental modification of Epic/Fortnite core assets.
---
### System & Safety (`category="System"`)
Advanced project diagnostics and protection layers.
| Tool Name | Description |
|---|---|
| `api_verse_get_schema` | Returns the Verse schema (props/events) for any class via digest parsing |
| `api_verse_refresh_schemas` | Forces a re-scan of all .digest.verse files |
| `system_build_verse` | Triggers a background UEFN build and scrapes for Verse errors |
| `system_get_last_build_log` | Reads the most recent UEFN log file for error diagnostics |
---
## Math & Performance Scaling
To prevent the editor from hanging during mass operations, Toolbelt uses
`unreal.ScopedSlowTask` — a native UE progress dialog with cancel support.
### Complexity Formula
For an operation across $A$ actors with $M$ material slots each:
$$\text{Complexity} = O(A + (A \times M))$$
### Chunked Execution Pattern
```python
# From core.py — wrap any iterable:
with with_progress(actors, "Applying materials") as bar_iter:
for actor in bar_iter:
_apply_preset_to_actor(actor, preset, "Chrome")
# ScopedSlowTask advances one step per actor
# User can cancel mid-way — no partial corrupt state
```
The progress bar displays a cancel button. Cancellation is handled gracefully; already-
processed actors retain their changes (inside the transaction, so all-or-nothing undo
still works at the outer level).
---
## Tool Reference
> **358 tools · 55 categories** — all return `{"status": "ok"/"error", ...}` structured dicts.
> Run any tool with `tb.run("tool_name", param=value)`.
> AI agents: use `tb.run("plugin_export_manifest")` to get the full machine-readable manifest.
**Quick navigation:**
[Actor Organization](#actor-organization-10) ·
[Alignment](#alignment-6) ·
[API Explorer](#api-explorer-10) ·
[Asset Management](#asset-management-5) ·
[Asset Tagger](#asset-tagger-6) ·
[Assets](#assets-11) ·
[Audio](#audio-4) ·
[Bulk Ops](#bulk-ops-9) ·
[Entities](#entities-2) ·
[Environmental](#environmental-2) ·
[Generative](#generative-2) ·
[Level Snapshot](#level-snapshot-8) ·
[Lighting](#lighting-6) ·
[Localization](#localization-2) ·
[Materials](#materials-10) ·
[MCP Bridge](#mcp-bridge-4) ·
[Measurement](#measurement-3) ·
[Optimization](#optimization-8) ·
[Pipeline](#pipeline-2) ·
[Post-Process](#post-process-4) ·
[Procedural](#procedural-12) ·
[Project](#project-7) ·
[Project Admin](#project-admin-3) ·
[Prop Patterns](#prop-patterns-9) ·
[Proximity Tools](#proximity-tools-6) ·
[Reference Auditor](#reference-auditor-7) ·
[Screenshot](#screenshot-4) ·
[Selection](#selection-3) ·
[Sequencer](#sequencer-2) ·
[Simulation](#simulation-2) ·
[Stamps](#stamps-5) ·
[System](#system-4) ·
[Tests](#tests-1) ·
[Text & Signs](#text--signs-14) ·
[Utilities](#utilities-12) ·
[Verse Helpers](#verse-helpers-26) ·
[Zone Tools](#zone-tools-7)
---
### Actor Organization (10)
| Tool | Description |
|---|---|
| `actor_attach_to_parent` | Last selected actor becomes parent of all others (Maya-style). Preserves world transforms. |
| `actor_detach` | Detach all selected actors from their parent. Preserves world transforms. |
| `actor_folder_list` | List all World Outliner folders with actor counts. |
| `actor_match_transform` | Copy the first selected actor's transform to all others. Toggle location/rotation/scale independently. |
| `actor_move_to_folder` | Move all selected actors into a named World Outliner folder (auto-created). |
| `actor_move_to_root` | Move all selected actors out of any folder to the root level. |
| `actor_rename_folder` | Rename a World Outliner folder — re-paths all actors inside it. Supports dry_run. |
| `actor_select_by_class` | Select all level actors whose class name contains a filter string. |
| `actor_select_by_folder` | Select all actors in a named World Outliner folder. |
| `actor_select_same_folder` | Expand selection to all actors sharing the first actor's folder. |
---
### Alignment (6)
| Tool | Description |
|---|---|
| `align_to_reference` | Align all selected actors' X, Y, or Z to the first or last actor's position. |
| `align_to_surface` | Snap selected actors' Z down to the nearest floor surface (uses line traces). |
| `align_to_grid_two_points` | Define a local XY grid from two anchor actors, then snap everything else to it. |
| `distribute_with_gap` | Set an exact cm gap between bounding boxes along an axis (not pivot-to-pivot). |
| `match_spacing` | Evenly space pivots between first and last actor — endpoints stay fixed. |
| `rotate_around_pivot` | Orbit selection around center-of-bounds or first actor by angle_deg on any axis. |
---
### API Explorer (10)
| Tool | Description |
|---|---|
| `api_search` | Fuzzy-search class/function names across the live `unreal` module. |
| `api_inspect` | Print full signature + docstring for any `unreal.*` class or function. |
| `api_list_subsystems` | Print every `*Subsystem` class available in this UEFN build. |
| `api_generate_stubs` | Write `.pyi` stub for one class (or all if name omitted). |
| `api_export_full` | Export a monolithic `unreal.pyi` stub for full IDE autocomplete. |
| `api_crawl_selection` | Deep-introspect selected actors, components, and properties → JSON. |
| `api_crawl_level_classes` | Headless map of all unique classes + exposed properties in the current level. |
| `api_sync_master` | One-click: level crawl + Verse schema merge → updates `docs/DEVICE_API_MAP.md`. |
| `device_catalog_scan` | Scan the Asset Registry for every Creative/Verse device available in Fortnite. |
| `world_state_export` | Export full live state of every actor (transforms + device properties) → `world_state.json`. |
---
### Asset Management (5)
| Tool | Description |
|---|---|
| `prefab_migrate_open` | Open the Prefab Asset Migrator — paste Ctrl+C T3D text to extract + copy all asset dependencies to another project. |
| `prefab_parse_refs` | Headless: parse a T3D prefab string and return all `/Game/` asset paths found. |
| `prefab_resolve_deps` | Headless: walk the AssetRegistry dependency graph for a list of package paths. |
| `prefab_export_to_disk` | Headless: copy resolved `.uasset` files from Content folder to a destination directory. |
| `ui_icon_import_open` | Open the UI Icon Importer — paste any image from clipboard (browser, Figma, Photoshop) to import as a UEFN Texture2D. |
---
### Asset Tagger (6)
All tags use the `TB:` namespace prefix to avoid collisions.
| Tool | Description |
|---|---|
| `tag_add` | Add a searchable metadata tag to all selected Content Browser assets. |
| `tag_remove` | Remove a tag from all selected Content Browser assets. |
| `tag_show` | Print all Toolbelt tags on selected assets. |
| `tag_search` | Find assets by tag key/value across a folder. |
| `tag_list_all` | List every unique tag key used under a folder with asset counts. |
| `tag_export` | Export full tag → asset index to `Saved/UEFN_Toolbelt/tag_export.json`. |
---
### Assets (11)
| Tool | Description |
|---|---|
| `rename_dry_run` | Preview naming convention violations without changing anything. |
| `rename_enforce_conventions` | Rename assets to follow Epic's prefix conventions (`SM_`, `T_`, `M_`, etc.). |
| `rename_strip_prefix` | Strip a specific prefix from all assets in a folder. |
| `rename_report` | Export a full naming audit JSON for a folder. |
| `organize_assets` | Sort assets into typed sub-folders by asset class. |
| `import_fbx` | Import a single FBX file into the Content Browser. |
| `import_fbx_folder` | Batch-import all FBX files in a folder. |
| `lod_audit_folder` | Report which static meshes in a folder have no LODs or collision. |
| `lod_auto_generate_selection` | Auto-generate LODs on selected actors' meshes. ⚠️ Disabled in UEFN (Quirk #18). |
| `lod_auto_generate_folder` | Auto-generate LODs on all meshes in a folder. ⚠️ Disabled in UEFN (Quirk #18). |
| `lod_set_collision_folder` | Batch-set collision complexity on all static meshes in a folder. |
---
### Audio (4)
| Tool | Description |
|---|---|
| `audio_place` | Spawn an AmbientSound actor at the camera position. Optionally assign a sound asset. |
| `audio_set_volume` | Set volume multiplier on all selected AmbientSound actors. |
| `audio_set_radius` | Override attenuation falloff radius on selected AmbientSound actors. |
| `audio_list` | List all AmbientSound actors in the level with label, location, and folder. |
---
### Bulk Ops (9)
| Tool | Description |
|---|---|
| `bulk_align` | Align all selected actors to the first actor's position on one axis. |
| `bulk_distribute` | Space selected actors evenly between the two extremes along an axis. |
| `bulk_randomize` | Apply random location / rotation / scale offsets to selected actors. |
| `bulk_snap_to_grid` | Snap all selected actors' locations to a world grid. |
| `bulk_reset` | Reset rotation to (0,0,0) and scale to (1,1,1) on all selected actors. |
| `bulk_mirror` | Mirror selected actors across an axis (reflects position about center). |
| `bulk_normalize_scale` | Set all selected actors to a uniform target scale. |
| `bulk_stack` | Stack selected actors vertically (Z) on top of each other. |
| `bulk_face_camera` | Rotate all selected actors to face the editor viewport camera (yaw only). |
---
### Entities (2)
| Tool | Description |
|---|---|
| `entity_list_kits` | List all available Standard Kits for the quick-spawn tool. |
| `entity_spawn_kit` | Spawn a pre-configured Standard Kit of Creative devices in one click. |
---
### Environmental (2)
| Tool | Description |
|---|---|
| `foliage_convert_selected_to_actor` | Convert selected StaticMeshActors into Foliage Actors. |
| `foliage_audit_brushes` | Audit all foliage brushes and return the mesh paths they use. |
---
### Generative (2)
| Tool | Description |
|---|---|
| `text_render_texture` | Render a text string into a transparent Texture2D asset natively in-editor. |
| `text_voxelize_3d` | Voxelize a text string into a 3D StaticMesh using procedural cubes. |
---
### Level Snapshot (8)
| Tool | Description |
|---|---|
| `snapshot_save` | Save a named JSON snapshot of actor transforms in the current level. |
| `snapshot_restore` | Restore actor transforms from a named snapshot (fully undoable). |
| `snapshot_list` | List all saved snapshots with actor count and timestamp. |
| `snapshot_diff` | Compare two snapshots — see what actors were added, removed, or moved. |
| `snapshot_compare_live` | Compare a saved snapshot against the current live level state. |
| `snapshot_export` | Copy a snapshot JSON to a custom path for sharing. |
| `snapshot_import` | Import a snapshot JSON from an external path. |
| `snapshot_delete` | Delete a saved snapshot by name. |
```python
tb.run("snapshot_save", name="before_scatter")
# ... make changes ...
tb.run("snapshot_compare_live", name="before_scatter")
tb.run("snapshot_restore", name="before_scatter")
```
---
### Lighting (6)
| Tool | Description |
|---|---|
| `light_place` | Spawn a light at camera position. `light_type`: point/spot/rect/directional/sky. |
| `light_set` | Batch-set intensity, color, and attenuation on selected lights. Only provided params change. |
| `light_list` | List all light actors with type, label, and location. |
| `sky_set_time` | Simulate time-of-day by pitching the DirectionalLight. `hour`: 0–24. |
| `light_cinematic_preset` | Apply mood preset to directional light + fog. Presets: Star Wars, Cyberpunk, Vibrant. |
| `light_randomize_sky` | Randomise sun pitch/yaw for quick look-dev iterations. |
---
### Localization (2)
| Tool | Description |
|---|---|
| `text_export_manifest` | Export all level text (TextRenderActors + devices) to CSV or JSON for translation. |
| `text_apply_translation` | Read a translated manifest and update level actors with new strings. |
---
### Materials (10)
**17 built-in presets:** `chrome` · `gold` · `neon` · `hologram` · `lava` · `ice` · `concrete` · `wood_oak` · `metal_rust` · `glass` · `team_red` · `team_blue` · `team_green` · `team_yellow` · `glow_pulse` · `scanlines` · `iridescent`
| Tool | Description |
|---|---|
| `material_list_presets` | Print all presets (17 built-in + custom). |
| `material_apply_preset` | Apply a named preset to all selected actors. |
| `material_randomize_colors` | Random HSV color on each selected actor. |
| `material_gradient_painter` | World-space color gradient across selection (`axis`, `color_a`, `color_b`). |
| `material_team_color_split` | Auto Red/Blue split by world X midpoint. |
| `material_pattern_painter` | Checkerboard or stripe alternation across selection. |
| `material_glow_pulse_preview` | Set emissive peak intensity for viewport preview. |
| `material_color_harmony` | Apply complementary / triadic / analogous palette to selection. |
| `material_save_preset` | Save current material params as a named reusable preset. |
| `material_bulk_swap` | Replace one material slot with another across selected actors or the full level. |
---
### MCP Bridge (4)
Start the listener in UEFN, then Claude Code (or any MCP client) can run all 358 tools by name, spawn actors, execute arbitrary Python, and more.
| Tool | Description |
|---|---|
| `mcp_start` | Start the HTTP listener so Claude Code can control UEFN directly. |
| `mcp_stop` | Stop the listener. |
| `mcp_restart` | Restart after hot-reload or port conflict. |
| `mcp_status` | Print port, running state, and command count. |
```python
tb.run("mcp_start")
# Output Log: [MCP] ✓ Listener running on http://127.0.0.1:8765
```
---
### Measurement (3)
| Tool | Description |
|---|---|
| `measure_distance` | 3D distance between two actors or total chain length of a selection. |
| `measure_travel_time` | Travel time between selected points at Walk/Run/Sprint/Vehicle speeds. |
| `spline_measure` | Precise world-space length of a selected spline actor. |
---
### Optimization (8)
| Tool | Description |
|---|---|
| `rogue_actor_scan` | Find actors with extreme scale, at-origin placement, or unnamed labels. |
| `memory_scan` | Full island memory scan — textures, meshes, sounds → JSON report. |
| `memory_scan_textures` | Audit textures and flag oversized ones (`warn_px`, `crit_px` thresholds). |
| `memory_scan_meshes` | Audit static meshes — polygon count, LODs, collision. |
| `memory_top_offenders` | Print the top N heaviest assets by estimated memory cost. |
| `memory_autofix_lods` | Auto-generate LODs on all meshes missing them in a folder. |
| `convert_to_hism` | Merge selected actors sharing the same mesh into a single HISM actor (one draw call). |
| `material_parent_audit` | Audit material instances — group by parent, flag orphaned MIs. |
---
### Pipeline (2)
| Tool | Description |
|---|---|
| `import_image_from_clipboard` | Capture the current Windows Clipboard image and import it as a Texture2D. |
| `import_image_from_url` | Download an image from a URL and import it into the Content Browser. |
---
### Post-Process (4)
| Tool | Description |
|---|---|
| `postprocess_spawn` | Find or create a global (infinite-extent) PostProcessVolume — no duplicates. |
| `postprocess_set` | Set bloom, exposure, contrast, vignette, saturation. Only provided params change. |
| `postprocess_preset` | Apply a named visual preset. Presets: `cinematic` · `night` · `vibrant` · `bleach` · `horror` · `fantasy` · `reset`. |
| `world_settings_set` | Change gravity (cm/s²) and time dilation world-wide. |
---
### Procedural (12)
| Tool | Description |
|---|---|
| `arena_generate` | Generate a symmetrical Red vs Blue arena. `size`: small/medium/large. |
| `scatter_props` | Scatter StaticMesh props in a radius with Poisson-disk distribution. |
| `scatter_hism` | Dense instanced scatter using HISM — one draw call for thousands of props. |
| `scatter_avoid` | Scatter props but skip positions within `avoid_radius` of matching actors. |
| `scatter_road_edge` | Place props along both shoulders of a path (road edges, forest borders). |
| `scatter_along_path` | Scatter prop clusters along a list of world-space path points. |
| `scatter_export_manifest` | Export positions/rotations of all scatter actors in a folder to JSON. |
| `scatter_clear` | Delete all scatter actors in a named World Outliner folder. |
| `spline_place_props` | Place props along a selected spline actor. |
| `spline_clear_props` | Delete all actors in the SplineProps folder. |
| `procedural_volume_scatter` | Scatter random meshes within a spherical or cubic boundary. |
| `procedural_wire_create` | Draw a procedural sagging wire/cable between exactly two selected actors. |
---
### Project (7)
| Tool | Description |
|---|---|
| `scaffold_list_templates` | List all available scaffold templates (built-in + saved custom). |
| `scaffold_preview` | Preview what a template would create — zero changes made. |
| `scaffold_generate` | Generate a professional Content Browser folder tree from a template. |
| `scaffold_save_template` | Save a custom folder list as a named reusable template. |
| `scaffold_delete_template` | Delete a saved custom template. |
| `scaffold_organize_loose` | Move loose `/Game` root assets into the project folder structure. |
| `organize_smart_categorize` | Auto-organize assets functionally (Meshes/Trees, Materials/Master, etc.) via smart keyword matching. |
**Built-in templates:** `uefn_standard` · `competitive_map` · `solo_dev` · `verse_heavy`
---
### Project Admin (3)
| Tool | Description |
|---|---|
| `project_setup` | One-command project setup: scaffold folders + deploy Verse game manager skeleton. |
| `system_backup_project` | Create a timestamped `.zip` backup of the Content folder. |
| `system_perf_audit` | Fast performance check of the current level. |
---
### Prop Patterns (9)
All pattern tools take `asset_path` + layout params and spawn at camera position.
| Tool | Description |
|---|---|
| `pattern_grid` | N×M rectangular grid (`rows`, `cols`, `spacing`). |
| `pattern_circle` | Evenly-spaced ring (`count`, `radius`). |
| `pattern_arc` | Partial arc between two angles (`count`, `radius`, `angle_start`, `angle_end`). |
| `pattern_spiral` | Archimedean spiral — radius grows with angle (`count`, `turns`). |
| `pattern_line` | Props between two world points (`count`, `start`, `end`). |
| `pattern_wave` | Sine-wave path (`count`, `amplitude`). |
| `pattern_helix` | 3D corkscrew (`count`, `height`, `radius`). |
| `pattern_radial_rows` | Concentric rings — dartboard/stadium layout (`rings`, `count_per_ring`). |
| `pattern_clear` | Delete actors spawned by pattern tools (`preview_only=True` to preview first). |
---
### Proximity Tools (6)
| Tool | Description |
|---|---|
| `actor_place_next_to` | Move last selected actor flush against the first on a specified face (`direction`: +X/-X/+Y/-Y/+Z/-Z). |
| `actor_chain_place` | Arrange selected actors end-to-end along an axis using bounds. |
| `actor_duplicate_offset` | Duplicate selected actor N times with cumulative offset — build arrays without copy-paste. |
| `actor_replace_class` | Replace every actor matching a class filter with a new asset. Always `dry_run=True` first. |
| `actor_cluster_to_folder` | Group nearby actors into World Outliner subfolders by XY-proximity radius. |
| `actor_copy_to_positions` | Stamp copies of the selected actor at every `[[x,y,z], ...]` in a list. |
---
### Reference Auditor (7)
Always `dry_run=True` before any destructive operation.
| Tool | Description |
|---|---|
| `ref_audit_orphans` | Find assets with no referencers — candidates for deletion. |
| `ref_audit_redirectors` | Find stale ObjectRedirectors left behind after moves/renames. |
| `ref_audit_duplicates` | Find assets sharing the same base name in different folders. |
| `ref_audit_unused_textures` | Find Texture2D assets not referenced by any material. |
| `ref_fix_redirectors` | Consolidate ObjectRedirectors. `dry_run=True` first. |
| `ref_delete_orphans` | Delete unreferenced assets. `dry_run=True` first. NOT undoable. |
| `ref_full_report` | Run all scans and export a JSON health report for the project. |
---
### Screenshot (4)
Output: `Saved/UEFN_Toolbelt/screenshots/{name}_{YYYYMMDD_HHMMSS}_{W}x{H}.png`
| Tool | Description |
|---|---|
| `screenshot_take` | Capture the viewport at any resolution (`width`, `height`, `name`). |
| `screenshot_focus_selection` | Auto-frame the selection, capture, then restore the camera. |
| `screenshot_timed_series` | Take N screenshots at a timed interval for before/after comparisons. |
| `screenshot_open_folder` | Print the screenshots output folder path to the Output Log. |
---
### Selection (3)
| Tool | Description |
|---|---|
| `select_by_property` | Select actors where a property matches a specific value. |
| `select_by_verse_tag` | Select actors that have a specific Verse tag. |
| `select_in_radius` | Select all actors of a class within a radius of the current selection. |
---
### Sequencer (2)
| Tool | Description |
|---|---|
| `seq_actor_to_spline` | Animate a selected actor along a selected spline path in the Sequencer. |
| `seq_batch_keyframe` | Add a transform keyframe for all selected actors at the current time. |
---
### Simulation (2)
| Tool | Description |
|---|---|
| `sim_generate_proxy` | Generate a Python simulation proxy for a selected Verse device. |
| `sim_trigger_method` | Trigger a discoverable method on a Verse device via the Python API. |
---
### Stamps (5)
Save a group of placed actors as a reusable named stamp and re-place it anywhere. Not the same as `prefab_migrate_open` — stamps are for level layout reuse, not asset migration.
| Tool | Description |
|---|---|
| `stamp_save` | Save selected StaticMesh actors as a named stamp → `Saved/UEFN_Toolbelt/stamps/{name}.json`. |
| `stamp_place` | Spawn stamp at camera (or `location=[x,y,z]`). `yaw_offset` rotates group; `scale_factor` rescales. |
| `stamp_list` | List all saved stamps with actor counts. |
| `stamp_info` | Print actor names, mesh paths, and relative offsets for a stamp. |
| `stamp_delete` | Delete a saved stamp by name. |
```python
tb.run("stamp_save", name="guard_post")
tb.run("stamp_place", name="guard_post", yaw_offset=90.0)
# Instant symmetric layout — 4 compass points
for angle, x, y in [(0,5000,0),(90,0,5000),(180,-5000,0),(287,0,-5000)]:
tb.run("stamp_place", name="guard_post", location=[x,y,0], yaw_offset=angle)
```
---
### System (4)
| Tool | Description |
|---|---|
| `system_build_verse` | Trigger Verse compilation and parse errors back as structured JSON. |
| `system_get_last_build_log` | Read the last 100 lines of the UEFN log for error analysis. |
| `system_optimize_background_cpu` | Disable UEFN sleep when alt-tabbed so Python/AI runs at max speed. |
| `verse_patch_errors` | AI error loop: read build log, extract errors + file content, return everything Claude needs to fix and redeploy. |
| `toolbelt_activity_log` | Rolling log of every `tb.run()` call — tool name, status, duration (ms), timestamp, error message. Newest first. |
| `toolbelt_activity_stats` | Aggregate session stats: total calls, error rate, slowest tool, most-called tool, last 5 errors. |
| `toolbelt_activity_clear` | Clear the in-memory ring buffer and the persisted `activity_log.json`. |
---
### Tests (1)
| Tool | Description |
|---|---|
| `toolbelt_integration_test` | ⚠️ Invasive — spawns/deletes actors. Run in a clean template level only. |
---
### Text & Signs (14)
| Tool | Description |
|---|---|
| `text_place` | Place a single styled 3D text actor at a world location. |
| `text_label_selection` | Auto-label every selected actor with its own name, floating above it. |
| `text_paint_grid` | Paint a coordinate grid of labels across a map area (A1, B2, …). |
| `text_color_cycle` | Place text actors cycling through a color palette. |
| `text_save_style` | Save a named text style preset for reuse across levels. |
| `text_list_styles` | Print all saved text style presets. |
| `text_clear_folder` | Delete all TextRenderActors in the Toolbelt text folder. |
| `sign_spawn_bulk` | Spawn N text signs in a row, column, or grid at a given location. |
| `sign_batch_edit` | Edit text/color/size on all selected signs — only provided fields change. |
| `sign_batch_set_text` | Assign individual text strings to each selected sign in order. |
| `sign_batch_rename` | Rename selected signs sequentially: `prefix_01`, `prefix_02`, … |
| `sign_list` | List all TextRenderActors in a folder with their current text. |
| `sign_clear` | Delete all signs in a named folder. |
| `label_attach` | Spawn a floating text label above each selected actor, parented so it follows movement. |
---
### Utilities (12)
| Tool | Description |
|---|---|
| `config_list` | Show all Toolbelt config keys — current values + whether overridden. |
| `config_get` | Read one config key. |
| `config_set` | Persist a config value (`key`, `value`). Survives restarts and updates. |
| `config_reset` | Reset a key (or `"all"`) back to built-in defaults. |
| `theme_list` | List all available Toolbelt UI themes. |
| `theme_get` | Get the currently active theme name. |
| `theme_set` | Switch the Toolbelt UI theme — applies live, persists across restarts. |
| `level_health_report` | Run all audit tools and return a unified 0–100 health score with A+/A/B/C/D/F grade. MCP-friendly. |
| `level_health_open` | Print a formatted Level Health Report to the Output Log (score + per-category breakdown). |
| `plugin_validate_all` | Validate all registered tools against Toolbelt schema requirements. |
| `plugin_list_custom` | List all loaded third-party custom plugins from `Saved/UEFN_Toolbelt/Custom_Plugins/`. |
| `plugin_export_manifest` | Export `tool_manifest.json` — full machine-readable index of all tools with parameter signatures. |
---
### Verse Helpers (26)
| Tool | Description |
|---|---|
| `verse_list_devices` | List all Verse/Creative device actors in the current level. |
| `verse_select_by_name` | Select all devices whose label contains a filter string. |
| `verse_select_by_class` | Select all devices of a given class name substring. |
| `verse_bulk_set_property` | Set a UPROPERTY on all selected Verse device actors at once. |
| `device_set_property` | AI write layer — set any base-class property on actors matching class/label/path filter. |
| `device_call_method` | V2 runtime control — call any exposed method on matched actors (`timer_start`, `Enable`, etc.). |
| `verse_export_report` | Export a JSON report of all Verse device properties to the Saved folder. |
| `verse_find_project_path` | Auto-detect the Verse source directory for the current UEFN project. |
| `verse_write_file` | AI deploy layer — write generated Verse directly into the project's Verse source directory. |
| `verse_gen_game_skeleton` | Generate a full `creative_device` game manager Verse stub. |
| `verse_gen_device_declarations` | Generate `@editable` device declarations from the current selection. |
| `verse_gen_elimination_handler` | Generate a Verse elimination event handler skeleton. |
| `verse_gen_scoring_tracker` | Generate a Verse zone scoring tracker device skeleton. |
| `verse_gen_prop_spawner` | Generate a trigger-controlled prop spawn/despawn Verse skeleton. |
| `verse_gen_custom` | Write arbitrary Verse to the snippets folder. |
| `verse_list_snippets` | List all generated snippet files in `Saved/UEFN_Toolbelt/snippets/`. |
| `verse_open_snippets_folder` | Open the Verse snippet library folder in Windows Explorer. |
| `verse_graph_open` | Interactive Verse Device Graph — force-directed layout, minimap, category filter, live sync, write-back. |
| `verse_graph_scan` | Headless scan → full device adjacency dict (MCP-friendly). |
| `verse_graph_export` | Export the device graph to JSON for external tooling. |
| `spline_to_verse_points` | Convert selected spline to a Verse `vector3` array literal. |
| `spline_to_verse_patrol` | Generate a full Verse patrol AI device skeleton from a selected spline. |
| `spline_to_verse_zone_boundary` | Convert selected spline to a Verse zone boundary + `IsPointInZone` helper. |
| `spline_export_json` | Export spline points to JSON (for `scatter_along_path` etc.). |
| `api_verse_get_schema` | Get the cached Verse schema for a device class. |
| `api_verse_refresh_schemas` | Refresh the Verse schema cache from current level devices. |
---
### Zone Tools (7)
Zone actors can be any box-shaped actor. **Convention:** select the zone actor first, then other actors.
| Tool | Description |
|---|---|
| `zone_spawn` | Spawn a visible cube zone marker at camera position in the "Zones" folder. |
| `zone_resize_to_selection` | Resize zone to exactly contain all other selected actors (+ `padding` cm). |
| `zone_snap_to_selection` | Move zone center to match combined selection bounds — no resize. |
| `zone_select_contents` | Select every level actor whose pivot is inside the zone's bounds. |
| `zone_move_contents` | Move zone + all actors inside it as a unit by a world-space offset. |
| `zone_fill_scatter` | Randomly scatter copies of an asset throughout the zone volume. |
| `zone_list` | List all actors in the "Zones" folder with center, width, depth, height. |
```python
tb.run("zone_spawn", width=4000, depth=4000, height=800, label="ArenaCenter")
tb.run("zone_resize_to_selection", padding=200)
tb.run("zone_select_contents")
tb.run("zone_move_contents", offset_x=5000)
tb.run("zone_fill_scatter", asset_path="/Engine/BasicShapes/Cube", count=50, min_spacing=300)
```
---
### Health Check
```python
# Run all 6 layers: Python env → UEFN API → Toolbelt core → MCP → Dashboard → Verse Book
tb.run("toolbelt_smoke_test")
# Results saved to Saved/UEFN_Toolbelt/smoke_test_results.txt
```
## Getting Started
> Tested on UEFN 40.00, Windows 11, Python 3.11.8 — March 2026.
### Step 1 — Enable Python in UEFN
Open your UEFN project. Go to **Edit → Project Settings**, search `python`, and tick:
- **Python Editor Script Plugin** ✓
> **Note:** UEFN only shows one Python checkbox — unlike full Unreal Engine which shows three. That's normal. Once you see "Enter a Python statement" appear in the Output Log, Python is active.
Restart UEFN when prompted.
---
### Step 2 — Install PySide6 (one-time, required for the dashboard UI)
Open a **regular Windows terminal** — not inside UEFN — and run:
```bat
"C:\Program Files\Epic Games\Fortnite\Engine\Binaries\ThirdParty\Python3\Win64\python.exe" -m pip install PySide6
```
> **Why outside UEFN?** UEFN's embedded Python can't run pip itself. You must use the same `python.exe` that UEFN uses, just run it directly from a terminal.
If the path above doesn't exist, find it at:
```
C:\Program Files\Epic Games\Fortnite\Engine\Binaries\ThirdParty\Python3\Win64\python.exe
```
You only need to do this once. PySide6 stays installed across projects.
---
### Step 3 — Install the Toolbelt into your UEFN project
**First time — run the installer:**
```bash
python install.py
```
It will find your Fortnite Projects folder automatically, let you pick a project, copy the Toolbelt in, and handle `init_unreal.py` safely — whether you already have one or not.
Or point it directly at a project:
```bash
python install.py --project "C:\Users\YOURNAME\Documents\Fortnite Projects\YOURPROJECT"
```
> **Updating after a `git pull`:** Re-run `python install.py` — it overwrites only the Toolbelt package, never your project's own files.
---
**If you're actively developing the Toolbelt itself — use `deploy.bat` instead:**
`deploy.bat` is the dev workflow tool. Double-click it or run it from a terminal. In addition to copying files it will:
- Check and install PySide6 automatically if it's missing
- Copy the `tests/` folder and `verse-book/` spec alongside the package
- Print the hot-reload command to paste into UEFN so you don't need a full restart
```bat
deploy.bat
```
> **Hot-reload after any code change** (no UEFN restart needed):
> ```python
> import sys; [sys.modules.pop(k) for k in list(sys.modules) if "UEFN_Toolbelt" in k]; import UEFN_Toolbelt as tb; tb.register_all_tools(); tb.launch_qt()
> ```
---
**Manual install** (if neither script works):
Your UEFN project lives at `C:\Users\YOURNAME\Documents\Fortnite Projects\YOURPROJECT\`.
```bat
xcopy /E /I /Y "PATH_TO_REPO\Content\Python\UEFN_Toolbelt" "C:\Users\YOURNAME\Documents\Fortnite Projects\YOURPROJECT\Content\Python\UEFN_Toolbelt"
```
Then copy `init_unreal.py` into `Content\Python\` — or if you already have one, paste only the package-discovery `for` loop from it into your existing file (under `# ── 2. Discover and load all packages`). It won't conflict with anything already there.
### Step 4 — Restart UEFN
Close and reopen your project. Watch the **Output Log**. You should see:
```
[TOOLBELT] ✓ All tools registered.
[TOOLBELT] ✓ Menu registered — look for 'Toolbelt' in the top menu bar.
```
A **Toolbelt** menu now appears in the top menu bar next to Help. No Blueprint setup, no manual script execution — it's automatic every time UEFN starts.
---
### Step 5 — Verify with the smoke test
In the Output Log, switch the input dropdown from **Cmd** to **Python** and run:
```python
import UEFN_Toolbelt as tb; tb.run("toolbelt_smoke_test")
```
> **UEFN Python console tip:** The console only accepts **one line at a time** — no multi-line pastes. Chain multiple statements on the same line using semicolons (`;`). A space before and after the semicolon works fine: `import UEFN_Toolbelt as tb; tb.run("toolbelt_smoke_test")`
Expected output in the log:
```
✓ All systems healthy — Toolbelt is ready.
```
| Layer | What It Verifies | Checks |
|---|---|---|
| **Layer 1** — Python Environment | stdlib, threading, sockets, HTTP server, file I/O | 13 |
| **Layer 2** — UEFN API Surface | `unreal` module, subsystems, AutomationLibrary, Materials | 13 |
| **Layer 3** — Toolbelt Core | All 76 modules loaded, 358 tools registered | 40 |
| **Layer 4** — MCP Bridge | 31 command handlers, HTTP listener state | 4 |
| **Layer 5** — Dashboard (PySide6) | PySide6 importable, QApplication, ToolbeltDashboard | 3 |
| **Layer 6** — Verse Book | clone present, git reachable, 22 chapters parsed | 12 |
| **Layer 7** — Integration | Fixture-based verification of context-aware tools — see Integration Testing section above | — |
> If you haven't installed PySide6 or cloned the verse-book yet, you may see a few non-critical failures.
> Complete Steps 2 and 7 in the Getting Started guide above to resolve them.
---
### Step 6 — Open the dashboard
After completing Step 2 (PySide6 installed), click **Toolbelt → Open Dashboard (PySide6)** in the top menu bar, or run:
```python
import UEFN_Toolbelt as tb; tb.launch_qt()
```
A dark-themed floating window opens with a left sidebar nav and 358 tools across 55 categories (including an About page).
**Two search modes:**
| Search | Where | What it does |
|---|---|---|
| **Find any tool…** | Sidebar (top) | Global — searches all 358 tools across every category by name, description, or tag. Results show a category badge so you know where each tool lives. |
| **Filter this page…** | Content header (top-right) | Within-category — hides/shows buttons on the current page as you type. Disappears during global search. |
> **If the dashboard doesn't open after installing PySide6:** The module may be cached from before PySide6 was installed. Paste this single line to clear and reload:
> ```python
> import sys; [sys.modules.pop(k) for k in list(sys.modules) if "UEFN_Toolbelt" in k]; import UEFN_Toolbelt as tb; tb.launch_qt()
> ```
---
### Step 7 — (Optional) Clone the Verse spec for AI codegen
In your repo folder, run:
```bat
git clone https://github.com/verselang/book.git verse-book
```
This gives Claude access to the full live Verse language spec for spec-accurate code generation. After cloning, the smoke test will show 63/64 (only PySide6 remaining if not installed).
---
### Step 8 — (Optional) Connect Claude Code via MCP
This lets Claude Code directly control UEFN — spawn actors, run any tool, generate Verse code — without you typing anything in the console.
**1. Install the MCP Python package** (regular terminal, not UEFN):
```bat
pip install mcp
```
**2. Create your `.mcp.json`:**
```bat
copy .mcp.json.template .mcp.json
```
Open `.mcp.json` and update the path to point to your repo:
```json
{
"mcpServers": {
"uefn-toolbelt": {
"command": "python",
"args": ["C:/Users/YOURNAME/Projects/UEFN-TOOLBELT/mcp_server.py"]
}
}
}
```
**3. Restart Claude Code** so it picks up the new server config.
**4. Start the listener inside UEFN** — paste this in the Python console:
```python
import UEFN_Toolbelt as tb; tb.run("mcp_start")
```
You should see in the Output Log:
```
[MCP] ✓ Listener running on http://127.0.0.1:8765
```
**5. Test it in Claude Code:**
```
run the toolbelt smoke test
```
Claude will call `run_toolbelt_tool("toolbelt_smoke_test")` through the bridge and show you the results.
> **Note:** The listener must be started in UEFN each session. You can also click **Dashboard → MCP → Start Listener** instead of pasting the command.
**What Claude can do once connected:**
- Run any of the 358 tools by name
- Spawn, move, delete actors directly
- Generate spec-accurate Verse code (pulls from the live verse-book spec)
- Execute arbitrary Python inside UEFN
- Batch multiple operations in one editor tick
---
### Step 9 — (Optional) Verse Snippet Library
The Verse tab in the dashboard has a built-in **Snippet Library** browser. Every time you generate a Verse snippet, it's saved to:
```
Saved/UEFN_Toolbelt/snippets/
├── game_systems/ ← game manager, elimination handler, scoring tracker
├── device_wiring/ ← @editable declarations, prop spawner
├── spline_tools/ ← patrol AI, zone boundary, vector3 arrays
└── custom/ ← your custom / Claude-generated snippets
```
In the dashboard:
1. Go to **Verse → Code Generation** and generate any snippet
2. Click **Refresh Library** in the Snippet Library section below
3. Your snippets appear organized by category — click any to **copy to clipboard**
4. Click **Open Folder** to browse them in Windows Explorer
5. Paste directly into your `.verse` file in UEFN
Snippets also auto-copy to the Windows clipboard when generated, so you can paste immediately without even opening the library.
---
### **The "Nuclear Reload" Command**
If you have modified the Toolbelt source code, run this in the UEFN console to refresh everything without a restart:
```python
import sys; [sys.modules.pop(k) for k in list(sys.modules) if "UEFN_Toolbelt" in k]; import UEFN_Toolbelt as tb; tb.register_all_tools(); tb.launch_qt()
```
> [!IMPORTANT]
> Always include `tb.register_all_tools()` after a pop, otherwise the tool registry will be empty!
### REPL Usage
> **UEFN console tip:** Chain multiple statements on one line with semicolons (`;`). The console does not accept multi-line pastes, but a single line can contain as many `;`-separated statements as you need.
**Dashboard & Health Check:**
```python
import UEFN_Toolbelt as tb; tb.launch_qt()
```
```python
import UEFN_Toolbelt as tb; tb.run("toolbelt_smoke_test")
```
**Force reload (after git pull or update):**
```python
import sys; [sys.modules.pop(k) for k in list(sys.modules) if "UEFN_Toolbelt" in k]; import UEFN_Toolbelt as tb; tb.launch_qt()
```
**Materials & Bulk Ops:**
```python
import UEFN_Toolbelt as tb; tb.run("material_apply_preset", preset="gold")
```
```python
import UEFN_Toolbelt as tb; tb.run("bulk_align", axis="Z")
```
```python
import UEFN_Toolbelt as tb; tb.run("arena_generate", size="large", apply_team_colors=True)
```
**Plugin Management:**
```python
import UEFN_Toolbelt as tb; tb.run("plugin_validate_all")
```
```python
import UEFN_Toolbelt as tb; tb.run("plugin_list_custom")
```
**API Explorer & Capability Crawling:**
```python
import UEFN_Toolbelt as tb; tb.run("api_crawl_level_classes")
```
```python
import UEFN_Toolbelt as tb; tb.run("api_crawl_selection")
```
```python
import UEFN_Toolbelt as tb; tb.run("api_export_full")
```
**Search & Discovery:**
```python
import UEFN_Toolbelt as tb; tb.registry.search("material")
```
```python
import UEFN_Toolbelt as tb; print(len(tb.registry.list_tools()))
```
**Handy Combos:**
```python
import UEFN_Toolbelt as tb; tb.run("toolbelt_smoke_test"); tb.run("plugin_validate_all")
```
```python
import UEFN_Toolbelt as tb; tb.run("snapshot_save", name="before"); tb.run("bulk_randomize", rot_range=360)
```
### Required Asset Setup (Material Master)
Create `Content/UEFN_Toolbelt/Materials/M_ToolbeltBase` in the material editor with
these five exposed parameters:
| Parameter | Type | Connects to |
|---|---|---|
| `BaseColor` | Vector | Base Color |
| `Metallic` | Scalar (0–1) | Metallic |
| `Roughness` | Scalar (0–1) | Roughness |
| `EmissiveColor` | Vector | × EmissiveIntensity → Emissive Color |
| `EmissiveIntensity` | Scalar (0–20) | (see above) |
Everything else works out of the box.
---
## Adding a New Tool
**3 steps. No boilerplate. No registration config.**
**Step 1** — Create `Content/Python/UEFN_Toolbelt/tools/my_tool.py`:
```python
"""UEFN TOOLBELT — My Tool"""
import unreal
from ..core import undo_transaction, require_selection, log_info
from ..registry import register_tool
@register_tool(
name="my_tool",
category="Utilities",
description="Does something amazing",
tags=["utility", "amazing"],
)
def run(**kwargs) -> dict:
actors = require_selection()
if actors is None:
return {"status": "error", "count": 0}
with undo_transaction("My Tool"):
for actor in actors:
log_info(f"Processing {actor.get_actor_label()}")
# your logic here
return {"status": "ok", "count": len(actors)}
```
> **Return contract:** Every tool must return a `dict` with a `"status"` key. This is what
> makes tools readable by AI agents via the MCP bridge — they read `result["status"]` and
> domain keys like `"count"` or `"path"` directly from the JSON response. Never return `None`.
**Step 2** — Add one line to `tools/__init__.py`:
```python
from . import my_tool
```
**Step 3** — Add a menu entry in `init_unreal.py`'s `_build_toolbelt_menu()` (optional):
```python
_add_entry(tb_sub, "Utilities", "MyTool",
"My Tool", "Does something amazing",
"import UEFN_Toolbelt as tb; tb.run('my_tool')")
```
Your tool is now registered, searchable, undo-safe, and accessible from the top menu bar.
---
## Plugin Hub & Community Ecosystem
UEFN Toolbelt ships a **live Plugin Hub** tab in the dashboard. Click **Refresh Hub** to pull the latest community registry directly from GitHub — no restart required.
### What you see in the Hub
| Section | What it shows |
|---|---|
| **Core Tools** (green, BUILT-IN badge) | 10 flagship modules built by Ocean Bennett — already included, View Source only |
| **Community Plugins** (blue, Install button) | Third-party tools submitted to `registry.json` — one-click download into `Custom_Plugins/` |
The registry is hosted at:
```
https://raw.githubusercontent.com/undergroundrap/UEFN-TOOLBELT/main/registry.json
```
### Getting your plugin listed
Add an entry to `registry.json` at the repo root and open a PR:
```json
{
"id": "my_cool_tool",
"name": "My Cool Tool",
"version": "1.0.0",
"author": "Your Name",
"author_url": "https://github.com/yourhandle",
"type": "community",
"description": "One sentence on what it does.",
"category": "Gameplay",
"tags": ["verse", "gameplay"],
"url": "https://github.com/yourhandle/yourrepo/blob/main/my_cool_tool.py",
"download_url": "https://raw.githubusercontent.com/yourhandle/yourrepo/main/my_cool_tool.py",
"min_toolbelt_version": "1.5.3",
"size_kb": 10
}
```
> `download_url` must be a raw GitHub URL. The dashboard downloads it directly into
> `Saved/UEFN_Toolbelt/Custom_Plugins/` — no zip, no installer, single `.py` file.
---
## Custom Plugins & Security
UEFN Toolbelt is a **security-first extensible platform**. Third-party developers can create custom tools that automatically load into the Dashboard and AI bridge — but the Toolbelt enforces strict sandboxing to protect your editor and project files.
### How Custom Plugins Work
Drop a `.py` file into `[Your Project]/Saved/UEFN_Toolbelt/Custom_Plugins/`. The Toolbelt will:
1. **Discover** the file on startup.
2. **Security-screen** it through four gates (see below).
3. **Register** it in the tool registry with a UI button and AI accessibility.
4. **Log** a tamper-evident audit record.
```python
# Example: one-file custom plugin
from UEFN_Toolbelt.registry import register_tool
from UEFN_Toolbelt import core
@register_tool(name="my_tool", category="Community", description="Does something cool")
def run(**kwargs) -> dict:
core.log_info("My custom tool ran!")
return {"status": "ok"} # always return a dict
```
### 🔒 Four-Gate Security Model
Every plugin passes through **four security gates** before it can execute inside your editor:
| Gate | What It Checks | What Happens If It Fails |
|---|---|---|
| **1. File Size Limit** | Rejects files > 50 KB | Blocks obfuscated or suspiciously large payloads |
| **2. AST Import Scanner** | Parses the source code *without executing it* using Python's `ast` module | Blocks dangerous imports: `subprocess`, `socket`, `ctypes`, `shutil`, `http`, `urllib`, `requests`, `webbrowser`, `smtplib`, `ftplib`, `xmlrpc`, `multiprocessing`, `signal`, `_thread` |
| **3. Namespace Protection** | Checks if the tool name collides with a core Toolbelt tool | Rejects registration — prevents hijacking of system commands like `toolbelt_smoke_test` |
| **4. SHA-256 Integrity Hash** | Computes a cryptographic fingerprint of every loaded plugin | Logged to `plugin_audit.json` — detect tampering by comparing hashes across sessions |
### 📋 Plugin Audit Log
Every time the Toolbelt boots, it writes a tamper-evident audit trail to:
```
Saved/UEFN_Toolbelt/plugin_audit.json
```
Each entry records the plugin name, its SHA-256 hash, file size, load status, and timestamp. If a plugin file changes between sessions, the hash changes — making unauthorized modifications instantly detectable.
**Read the full developer guide here:** [docs/plugin_dev_guide.md](docs/plugin_dev_guide.md)
---
## API Capability Crawler
Epic's documentation for the UEFN Python API is incomplete. Many Fortnite device properties are exposed to Python but never documented. The **Capability Crawler** solves this by brute-forcing introspection on live actors inside the editor — mapping out exactly what variables exist, which are readable, and what values they hold.
### Why This Matters
- **For Verse developers:** Discover hidden properties on Fortnite devices that aren't in the docs
- **For tool builders:** Know exactly what you can automate before writing a single line of code
- **For AI-assisted workflows:** Generate a machine-readable API map that Claude can analyze and act on
### Verified Results (Live)
Tested on a real UEFN project (March 2026):
```
Found 41 total actors, distilling to 12 unique classes...
✓ Level class schema saved: api_level_classes_schema.json
```
The crawler scanned 41 actors, identified 12 unique class types, and deep-introspected every property, method, and component hierarchy — all in under a second, with zero editor impact.
### Workflow
**Step 1 — Scan your level** (headless, no selection required):
```python
import UEFN_Toolbelt as tb; tb.run("api_crawl_level_classes")
```
**Step 2 — Scan a specific selection** (select actors in viewport first):
```python
import UEFN_Toolbelt as tb; tb.run("api_crawl_selection")
```
**Step 3 — Review the output:**
```
Saved/UEFN_Toolbelt/api_level_classes_schema.json ← full level scan
Saved/UEFN_Toolbelt/api_selection_crawl.json ← selection scan
```
### Output Format
The JSON contains a per-class schema with every exposed property, its type, readability, and an example value:
```json
{
"classes": {
"StaticMeshActor": {
"properties": {
"root_component": { "type": "SceneComponent", "readable": true },
"mobility": { "type": "str", "readable": true, "example_value": "EComponentMobility.STATIC" }
},
"methods": ["get_actor_label", "set_actor_location", "get_components_by_class"],
"components": {
"StaticMeshComponent0": {
"properties": { "static_mesh": { "type": "StaticMesh", "readable": true } }
}
}
}
}
}
```
Properties marked `"readable": false` with an `"error"` field indicate locked-down C++ internals that Epic has not yet exposed.
### AI-Powered Analysis (Claude Integration)
Because the crawler output is structured JSON, it becomes a direct input for AI analysis. After running a crawl, you can ask Claude Code:
> *"Read `Saved/UEFN_Toolbelt/api_level_classes_schema.json` and tell me which properties on each device class are writable from Python."*
> *"Which classes in my level have components with exposed transform properties I could animate?"*
> *"Generate a Verse class that references every device in my level using the correct variable types from the crawl data."*
The crawler is the **data collection engine**. Claude is the **analysis layer**. Together they let you reverse-engineer Epic's undocumented API surface without ever reading a single page of docs.
---
## Fortnite Device API Mapping
The Toolbelt includes a fully schema-derived API map built from real UEFN level introspection —
not guesswork. The reference schema (`uefn_reference_schema.json`) was generated by running
`api_crawl_level_classes` against a live level with 76 actors, producing a 1.6MB JSON
with every readable and restricted property across 14 C++ classes.
### Discovery Workflow
```python
import UEFN_Toolbelt as tb
# Crawl your level — produces docs/api_level_classes_schema.json
tb.run("api_crawl_level_classes")
# Or use the Dashboard "Sync Level Schema" button for one-click sync
```
### The Three Schema Layers
| Layer | File | What it contains |
|:---|:---|:---|
| **Tool Layer** | `Saved/UEFN_Toolbelt/tool_manifest.json` | 358 tools — names, params, types, defaults |
| **C++ Layer** | `docs/uefn_reference_schema.json` | 14 actor classes, 1,031 properties with types + defaults |
| **Verse Layer** | `docs/api_level_classes_schema.json` | Your project's `@editable` device properties (git-ignored) |
### Verified Actor Classes (from real schema)
| Class | Readable Props | Key Automation Properties |
|:---|:---|:---|
| `BuildingProp` | 130 | `is_invulnerable`, `no_collision`, `team_index`, `static_mesh`, `allow_interact` |
| `BuildingFloor` | 125 | `snap_grid_size` (512), `register_with_structural_grid`, `resource_type` |
| `FortCreativeDeviceProp` | 130 | Verse shell — use `getattr()` or `verse_bulk_set_property` |
| `FortPlayerStartCreative` | 41 | `is_enabled`, `applicable_team`, `priority_group`, `use_as_island_start` |
| `FortMinigameSettingsBuilding` | 53 | `max_players`, `join_in_progress_behavior`, `creative_matchmaking_privacy` |
| `FortStaticMeshActor` | 31 | `static_mesh_component`, non-destructible by default |
| `TextRenderActor` | 31 | `text_render` component — world text placed by `text_place` tool |
| `FortWaterBodyActor` | 37 | `water_body_primary_color`, `water_waves`, `overlapping_players` |
| `Actor` (base) | 30 | `hidden`, `tags`, `replicates`, `net_dormancy`, `pivot_offset` |
> **Full reference** — property tables, enums, collision groups, automation patterns, and
> access cheatsheet: **[docs/DEVICE_API_MAP.md](docs/DEVICE_API_MAP.md)**
>
> **Creative device reference** — Trigger, Score Manager, Teleporter, Guard Spawner, channel
> wiring, Verse code gen: **[docs/FORTNITE_DEVICES.md](docs/FORTNITE_DEVICES.md)**
---
## PySide6 Dashboard
The real dashboard. One `pip install PySide6` (or just run `deploy.bat`) and you get a full floating window with a scrollable left sidebar nav, two search modes, dark theme, inline parameter inputs, and live status feedback.
**No Blueprint. No UMG. No Restart.**
### What you get
```
⬡ UEFN Toolbelt
┌─────────────────────────────────────────────────────┐
│ Materials │ Procedural │ Bulk Ops │ Text │ Assets │ Verse │ Project │ API │
├─────────────────────────────────────────────────────┤
│ ┌─ PRESETS ──────────────────────────────────────┐ │
│ │ [Chrome] [Gold] [Neon] [Hologram] │ │
│ │ [Lava] [Plasma][Ice] [Mirror] │ │
│ │ [Wood] [Conc.] [Rubber][Carbon] │ │
│ │ [Jade] [Obsid.][Rusty] [Team Red][Team Blue] │ │
│ └────────────────────────────────────────────────┘ │
│ ┌─ OPERATIONS ───────────────────────────────────┐ │
│ │ [Randomize Colors ] │ │
│ │ [Team Color Split (Red / Blue by X pos) ] │ │
│ │ [Color Harmony] [triadic ▾] │ │
│ │ [Gradient Painter (Blue → Red along X) ] │ │
│ └────────────────────────────────────────────────┘ │
│ ────────────────────────────────────────────────── │
│ ✓ material_apply_preset — done 4s │
└─────────────────────────────────────────────────────┘
```
### Launch
```python
import UEFN_Toolbelt as tb; tb.launch_qt()
```
Or: **Toolbelt → Open Dashboard (PySide6)** from the top menu bar.
### Color reference (QSS applied automatically)
| Element | Color |
|---|---|
| Window background | `#181818` |
| Section / card background | `#212121` |
| Button default | `#262626` border `#363636` |
| Button hover | `#333333` |
| Button pressed / accent | `#3A3AFF` |
| Active tab | `#3A3AFF` white text |
| Body text | `#CCCCCC` |
| Section headers | `#555555` small-caps |
| Status: success | `#44FF88` |
| Status: error | `#FF4444` |
### Event loop integration
The dashboard registers a `slate_post_tick_callback` that calls
`QApplication.processEvents()` once per editor tick — the same pattern used in
[Kirch's uefn_listener.py](https://github.com/KirChuvakov/uefn-mcp-server).
This keeps the Qt window responsive while UEFN runs heavy operations without
needing a separate thread.
---
## Blueprint Dashboard (Legacy / Optional)
The Blueprint approach still works and is documented for completeness.
**Prefer the PySide6 dashboard** — it launches faster and requires no Blueprint setup.
The top-menu Toolbelt entries work without this. Build the widget if you want
a visual panel with buttons, sliders, and live text input instead of the REPL.
### Create the Widget Blueprint
1. In Content Browser → `Content/UEFN_Toolbelt/Blueprints/`
2. Right-click → **Editor Utilities → Editor Widget**
3. Name it exactly: `WBP_ToolbeltDashboard`
4. Double-click to open the UMG Designer.
### Dark Mode Styling
The dashboard targets a `#181818 / #212121` dark theme to match the UEFN editor chrome.
Apply these values in the **Details** panel for each widget:
| Element | Property | Value |
|---|---|---|
| Root Canvas / Overlay | **Background Color** | `#181818` (R 0.024, G 0.024, B 0.024, A 1.0) |
| Section panels / cards | **Background Color** | `#212121` (R 0.033, G 0.033, B 0.033, A 1.0) |
| Header Text Block | **Color and Opacity** | `#FFFFFF` / Alpha 1.0 |
| Body / label Text Blocks | **Color and Opacity** | `#CCCCCC` (R 0.8, G 0.8, B 0.8) |
| Tab bar active button | **Normal → Tint** | `#3A3AFF` (accent blue) |
| Tab bar inactive button | **Normal → Tint** | `#2A2A2A` |
| Tool buttons (default) | **Normal → Tint** | `#2D2D2D` |
| Tool buttons (hover) | **Hovered → Tint** | `#3D3D3D` |
| Tool buttons (pressed) | **Pressed → Tint** | `#1A1AFF` |
| Separator / divider | **Border** widget, **Brush Tint** | `#383838` |
| Warning / critical text | **Color and Opacity** | `#FF4444` |
| Success / ok text | **Color and Opacity** | `#44FF88` |
**Setting background color on a Border widget** (UMG has no plain colored panel):
1. Drag a **Border** widget into the hierarchy where you need a colored region.
2. In **Details → Brush → Image**: leave blank (or use a 1×1 white texture).
3. Set **Brush → Tint** to your hex color.
4. Set **Content → Horizontal/Vertical Alignment** to `Fill`.
5. Drop your content widgets *inside* the Border as children.
**Font recommendation:** Import `Rajdhani-SemiBold.ttf` (free, SIL OFL) as a Font Face asset
at `/Game/UEFN_Toolbelt/Fonts/Rajdhani`. Set it on all Text Blocks for a modern HUD look.
Fallback: the built-in `Roboto` works fine.
### Recommended UMG Layout
```
[Canvas Panel] — #181818 background (via full-screen Border)
└── [Vertical Box] (fill screen, padding 0)
├── [Border] header bar — #212121, height 48px
│ └── [Horizontal Box]
│ ├── [Text Block] "⬡ UEFN TOOLBELT" white, 18pt, Rajdhani
│ └── [Spacer]
├── [Border] tab bar — #181818, height 36px
│ └── [Horizontal Box]
│ Buttons: Materials | Procedural | BulkOps | Text | Optimize | Verse
│ (blue tint on active, dark on rest)
└── [Widget Switcher] (content area — fills remaining space)
└── [Scroll Box] (per tab)
└── [Vertical Box] (tool buttons, padding 8px, gap 4px)
└── [Border] per button row — #2D2D2D hover #3D3D3D
└── [Button] → Execute Python Command
```
**Tip:** Wire each tab button's **On Clicked** → set the **Active Widget Index** on the
Widget Switcher. Each index corresponds to one category's button panel.
**Tip:** Add a `[Text Block]` at the bottom of each panel wired to a Blueprint variable
`StatusText`. On tool button click, set `StatusText` to `"Running..."` before the
Execute Python Command node and `"Done ✓"` after. This gives live feedback without
needing to read the Output Log.
### Wiring a Button to a Tool
1. Add a **Button** widget with a **Text Block** child (e.g. "Apply Chrome").
2. In the **Graph** tab, find the button's **On Clicked** event.
3. Search for and add: **Execute Python Command**.
4. Wire **On Clicked** → **Execute Python Command**.
5. Type the tool call into the **Python Command** string pin:
```
import UEFN_Toolbelt as tb; tb.run("material_apply_preset", preset="chrome")
```
### Complete Button String Reference
```python
# Materials
import UEFN_Toolbelt as tb; tb.run("material_apply_preset", preset="chrome")
import UEFN_Toolbelt as tb; tb.run("material_apply_preset", preset="gold")
import UEFN_Toolbelt as tb; tb.run("material_apply_preset", preset="neon")
import UEFN_Toolbelt as tb; tb.run("material_apply_preset", preset="hologram")
import UEFN_Toolbelt as tb; tb.run("material_randomize_colors")
import UEFN_Toolbelt as tb; tb.run("material_gradient_painter", color_a="#0044FF", color_b="#FF2200")
import UEFN_Toolbelt as tb; tb.run("material_team_color_split")
import UEFN_Toolbelt as tb; tb.run("material_color_harmony", harmony="triadic")
# Procedural
import UEFN_Toolbelt as tb; tb.run("arena_generate", size="small")
import UEFN_Toolbelt as tb; tb.run("arena_generate", size="medium")
import UEFN_Toolbelt as tb; tb.run("arena_generate", size="large")
import UEFN_Toolbelt as tb; tb.run("spline_place_props", count=20, align_to_tangent=True)
import UEFN_Toolbelt as tb; tb.run("scatter_props", mesh_path="/Engine/BasicShapes/Sphere", count=100, radius=3000.0)
# Bulk Ops
import UEFN_Toolbelt as tb; tb.run("bulk_align", axis="Z")
import UEFN_Toolbelt as tb; tb.run("bulk_distribute", axis="X")
import UEFN_Toolbelt as tb; tb.run("bulk_randomize", rot_range=360.0, randomize_rot=True)
import UEFN_Toolbelt as tb; tb.run("bulk_snap_to_grid", grid=100.0)
import UEFN_Toolbelt as tb; tb.run("bulk_reset")
import UEFN_Toolbelt as tb; tb.run("bulk_mirror", axis="X")
# Text & Signs
import UEFN_Toolbelt as tb; tb.run("text_place", text="ZONE", color="#FFDD00", location=(0,0,200))
import UEFN_Toolbelt as tb; tb.run("text_label_selection", color="#00FFCC")
import UEFN_Toolbelt as tb; tb.run("text_paint_grid", cols=4, rows=4, cell_size=2000.0)
import UEFN_Toolbelt as tb; tb.run("text_clear_folder")
# Assets
import UEFN_Toolbelt as tb; tb.run("rename_dry_run", scan_path="/Game")
# lod_auto_generate_folder is disabled in UEFN (see UEFN_QUIRKS.md #18) — use lod_audit_folder to find meshes, then add LODs manually in the Static Mesh Editor
import UEFN_Toolbelt as tb; tb.run("lod_audit_folder", folder_path="/Game")
import UEFN_Toolbelt as tb; tb.run("organize_assets", source_path="/Game/Imports")
# Verse
import UEFN_Toolbelt as tb; tb.run("verse_gen_device_declarations")
import UEFN_Toolbelt as tb; tb.run("spline_to_verse_points")
import UEFN_Toolbelt as tb; tb.run("spline_to_verse_patrol")
import UEFN_Toolbelt as tb; tb.run("verse_list_devices")
import UEFN_Toolbelt as tb; tb.run("verse_bulk_set_property", property_name="bIsEnabled", value=True)
# Optimization
import UEFN_Toolbelt as tb; tb.run("memory_scan", scan_path="/Game")
import UEFN_Toolbelt as tb; tb.run("memory_scan_textures", scan_path="/Game")
import UEFN_Toolbelt as tb; tb.run("memory_scan_meshes", scan_path="/Game")
import UEFN_Toolbelt as tb; tb.run("memory_top_offenders", limit=20)
import UEFN_Toolbelt as tb; tb.run("memory_autofix_lods", scan_path="/Game/Meshes")
# API Explorer
import UEFN_Toolbelt as tb; tb.run("api_export_full")
import UEFN_Toolbelt as tb; tb.run("api_list_subsystems")
import UEFN_Toolbelt as tb; tb.run("api_search", query="material")
import UEFN_Toolbelt as tb; tb.run("api_inspect", name="EditorActorSubsystem")
```
### Adding a Text Input to a Button
When a tool needs a dynamic argument (file path, preset name, text string):
1. Add an **Editable Text Box** above the button.
2. In the button's **On Clicked** event, drag off the text box → **Get Text**.
3. **Convert Text to String** → feed into a **Format Text** node to build the command:
```
import UEFN_Toolbelt as tb; tb.run("text_place", text="{0}", location=(0,0,200))
```
4. Pass the formatted string into **Execute Python Command**.
### Opening the Dashboard Manually
```python
import UEFN_Toolbelt as tb; tb._try_open_widget()
```
Or pin it permanently: **Window → UEFN Toolbelt** (registers as a dockable tab).
---
## Community & Prior Art
This project builds on the first 48 hours of real-world community testing after the
UEFN Python v40.00 experimental drop (March 2026).
**API discovery vs. Execution: The "Superset" Architecture**
> **[KirChuvakov/uefn-mcp-server](https://github.com/KirChuvakov/uefn-mcp-server) · [@KirchCreator](https://x.com/KirchCreator)**
> Kirch pioneered the MCP bridge pattern for UEFN Python — the queue + Slate tick
> architecture that makes it possible to call `unreal.*` from an external process
> without deadlocking the editor. He built the foundational "API Ground Truth" tools
> (`dump_uefn_api.py`, `generate_uefn_stub.py`, `uefn_listener.py`). Full credit.
>
> **How UEFN Toolbelt differs:** Kirch's project focused on *Discovery* — mapping the API
> and teaching AI what exists. **UEFN Toolbelt is an Execution Engine** — 358 tools for
> actually building levels, built on top of that same bridge architecture. You don't need
> to run a standalone MCP server alongside Toolbelt — Toolbelt is the all-in-one superset.
**PySide6 UI approach inspired by:**
- Early PySide6 experiments — proving out polished Qt UIs in the UEFN Python space
- Community validations — confirming PySide6 install paths and event loop patterns
**Community requests that shaped this toolbelt:**
- Map Makers — Spline → Verse data converters (shipped as `spline_to_verse.py`)
- Device Programmers — Verse device property editing workarounds (shipped as `verse_device_editor.py`)
- Performance Testers — memory profilers, API explorers, level snapshots, prop patterns
---
## ⚠️ Critical: The "Main Thread Lock" Quirk
Because UEFN runs Python on the **main render thread**, there is a unique deadlock behavior you must avoid:
* **Async APIs**: Commands like `take_high_res_screenshot` or background asset saves are queued for the **next frame**.
* **The Deadlock**: If your Python script uses `time.sleep()` or a long loop to "wait" for a file to appear, you are **blocking the main thread**. Unreal cannot finish the current frame, so it will **never write the file** while your script is waiting.
* **The Solution**: Trigger the action and exit. Do not attempt to verify file existence in the same execution block if the engine needs to tick to produce that file.
## 🚀 Automated "Hands-Free" Verification
The UEFN Toolbelt includes a sophisticated integration test suite that validates tools **programmatically**.
* **How it works**: The `integration_test.py` script uses the Unreal Python API to remote-control the editor. It spawns actors, modifies transforms, applies tags, and captures screenshots—verified against the engine's internal state.
* **What it tests**: It validates the "Logic & Engine" layer (Layer 3). While it doesn't click the physical buttons on the Pyside6 UI, it proves that the toolbelt's brain is working perfectly.
* **Benefits**: Run 12 tests in under 5 seconds (excluding screenshots) to ensure no regressions were introduced.
---
## 🏎️ Optimized Development Workflow (Hot Reload)
You **never need to restart UEFN** when developing for the Toolbelt.
1. **Edit** your Python code in your IDE.
2. **Run** `deploy.bat` to sync files to your UEFN project.
3. **Paste** the right command from the table below into the UEFN Python console.
> [!WARNING]
> Always include `tb.register_all_tools()` after the module pop. If you skip it the registry will be empty and every tool call will fail with "Unknown Tool".
### Quick-Reference Command Table
Copy the command that matches what you're doing right now:
| What you're working on | Command to paste in UEFN console |
|---|---|
| **Standard dev — open dashboard** | `import sys; [sys.modules.pop(k) for k in list(sys.modules) if "UEFN_Toolbelt" in k]; import UEFN_Toolbelt as tb; tb.register_all_tools(); tb.launch_qt()` |
| **Verse Device Graph window** | `import sys; [sys.modules.pop(k) for k in list(sys.modules) if "UEFN_Toolbelt" in k]; import UEFN_Toolbelt as tb; tb.register_all_tools(); tb.run("verse_graph_open")` |
| **Run integration tests** | `import sys; [sys.modules.pop(k) for k in list(sys.modules) if "UEFN_Toolbelt" in k]; import UEFN_Toolbelt as tb; tb.register_all_tools(); tb.run("toolbelt_integration_test")` |
| **Quick smoke test only** | `import sys; [sys.modules.pop(k) for k in list(sys.modules) if "UEFN_Toolbelt" in k]; import UEFN_Toolbelt as tb; tb.register_all_tools(); tb.run("toolbelt_smoke_test")` |
| **Sync docs + Verse IQ** | `import sys; [sys.modules.pop(k) for k in list(sys.modules) if "UEFN_Toolbelt" in k]; import UEFN_Toolbelt as tb; tb.register_all_tools(); tb.run("api_sync_master")` |
| **Headless reload (no UI)** | `import sys; [sys.modules.pop(k) for k in list(sys.modules) if "UEFN_Toolbelt" in k]; import UEFN_Toolbelt as tb; tb.register_all_tools()` |
All of these also appear in `deploy.bat` output — after deploying, just copy from the terminal.
---
## 🛠 Troubleshooting
| Problem | Solution |
|---|---|
| `ModuleNotFoundError: UEFN_Toolbelt` | Files not copied correctly OR this is a new project. **UEFN only scans `Content/Python` on startup.** If you just deployed to a new project, restart UEFN. |
| `Qt dashboard is blank / doesn't open` | PySide6 not installed. Run `pip install PySide6` using the **UEFN Python exe** at `<UE_INSTALL>/Engine/Binaries/ThirdParty/Python3/Win64/python.exe`. |
| `ImportError: No module named PySide6` | Same as above. Pip installing into your system Python won't help — must use the UE Python binary. |
| Python settings not sticking | Enable Python in **Edit → Project Settings** (not Preferences) — UEFN only has **one** Python checkbox, not three like full UE. Tick it, restart. |
| Material instances not appearing in viewport | Missing `update_material_instance()` call — already fixed. Check `PARENT_MATERIAL_PATH` in `material_master.py` points to a real asset. |
| `spawn_actor_from_class returned None` | No level is open, or world context is unavailable. Open a level first. |
| Blueprint widget doesn't open | Create `WBP_ToolbeltDashboard` in `Content/UEFN_Toolbelt/Blueprints/` — or just use the PySide6 dashboard instead. |
| Property set fails on Verse device | Property may be read-only in-editor or the name differs. Check the Details panel spelling. |
| FBX import produces no assets | Check Output Log for FBX parse errors. Use forward slashes in file paths. |
| Scatter actors not visible | Check the mesh path is valid — tool falls back to `/Engine/BasicShapes/Cube` if it can't find the mesh. |
| Something isn't loading after an update | Module may be cached. Paste: `import sys; [sys.modules.pop(k) for k in list(sys.modules) if "UEFN_Toolbelt" in k]; import UEFN_Toolbelt as tb; tb.launch_qt()` |
| LOD generation fails | `StaticMeshEditorSubsystem.set_lods_with_notification` requires the mesh to be fully loaded. Run `load_asset()` first. |
| Menu bar entry missing (UEFN ≤ 40.10) | `init_unreal.py` failed silently — check Output Log for `[TOOLBELT]` error lines on startup. |
| Menu bar entry missing (UEFN 40.20+) | Known Epic platform restriction — `ToolMenus` extensions for third-party Python are sandboxed in 40.20. The Toolbelt menu will not appear regardless of configuration. Use `tb.launch_qt()` or the dashboard instead. All 358 tools and every other feature work normally. |
---
## Performance Tips
- **Slow on large selections?** `with_progress(actors, "label")` gives a cancellable progress bar — already used in all bulk tools.
- **Accumulating material instances?** They stack in `/Game/UEFN_Toolbelt/Materials/Instances/`. Periodically delete unused ones via the Content Browser.
- **Arena generation slow?** Increase `tile_size` in `ARENA_PRESETS` so fewer tiles are needed for the same area.
- **Dense scatter hanging the editor?** Use `scatter_hism` instead of `scatter_props` for counts above ~500 — it's one actor instead of N.
- **Spline sampling too slow?** Reduce `sample_count` in `spline_to_verse_points` — control points (default) are instant.
---
## MCP — Connect Any AI to UEFN
UEFN Toolbelt ships a full two-process MCP (Model Context Protocol) architecture so **any MCP-compatible AI** can directly control the editor — no copy-pasting, no manual Python runs.
**This works with any MCP-compatible AI — Claude Code, Cursor, Windsurf, Zed, Continue,
OpenClaw, NemoClaw, or any custom agent that speaks the MCP protocol.**
Claude is the recommended driver because of the verse-book spec integration and CLAUDE.md
context loading, but the bridge is completely model-agnostic. If your AI supports MCP, it connects.
**How it works:**
```
Your AI ←── MCP stdio ──→ mcp_server.py ←── HTTP ──→ UEFN editor
(IDE) (runs outside) (mcp_bridge.py)
```
**One-time setup:**
```bat
pip install mcp
```
**Start the listener in UEFN:**
```python
import UEFN_Toolbelt as tb; tb.run("mcp_start")
# Output Log: [MCP] ✓ Listener running on http://127.0.0.1:8765
```
`.mcp.json` is already in the repo root — any MCP client picks it up automatically.
**What any connected AI can do:**
- Run any of the 358 toolbelt tools by name (`run_toolbelt_tool`)
- Spawn, move, delete actors; read selected actors live
- List, rename, duplicate, import, delete Content Browser assets
- Execute arbitrary Python inside UEFN with full `unreal.*` access
- Create material instances with parameters
- Batch multiple operations in a single editor tick
- Undo / redo editor actions
- Read command history with per-command timing
- Control the viewport camera
- Generate spec-accurate Verse code (see below)
**External client — works with any AI, any language:**
`client.py` in the project root is a stdlib-only HTTP client that connects directly to the same listener. No MCP, no SDK, no dependencies — plain HTTP that works from anywhere:
```python
from client import ToolbeltClient
ue = ToolbeltClient()
ue.ping()
ue.run_tool("scatter_hism", count=200, radius=4000.0)
actors = ue.get_all_actors()
ue.batch([
{"command": "run_tool", "params": {"tool_name": "snapshot_save"}},
{"command": "save_current_level", "params": {}},
])
```
Because the bridge is plain HTTP + JSON, **any agent that can make an HTTP request can control UEFN** — you are not locked into Claude:
| Agent | How |
|---|---|
| **LM Studio** (Qwen, Llama, Mistral, etc.) | Point your local model's tool-calling at `client.py` or call `http://127.0.0.1:8765` directly |
| **Ollama** | Same — any model with function-calling support works |
| **curl / bash** | `curl -X POST http://127.0.0.1:8765 -d '{"command":"ping","params":{}}'` |
| **Go / Rust / Node** | Any HTTP client, no library needed |
| **CI pipelines** | Run `client.py` from GitHub Actions, Jenkins, whatever |
```bash
# Full UEFN control from curl — no Python, no SDK
curl -s -X POST http://127.0.0.1:8765 \
-H "Content-Type: application/json" \
-d '{"command":"run_tool","params":{"tool_name":"snapshot_save","kwargs":{}}}'
```
**Claude is the recommended driver** because of the verse-book spec integration, CLAUDE.md auto-context loading, and the best reasoning for complex autonomous tasks — but the bridge is fully model-agnostic. Confirmed compatible: Claude Code, Cursor, Windsurf, Zed, Continue, OpenClaw, NemoClaw (Linux/NVIDIA), or any custom agent built against the HTTP API directly.
MCP bridge architecture inspired by [Kirch's uefn-mcp-server](https://github.com/KirChuvakov/uefn-mcp-server) ([@KirchCreator](https://x.com/KirchCreator)) — full credit for the queue + Slate tick pattern.
---
## Spec-Accurate Verse Code Generation
Every other Verse code generator on the market generates from static templates written once and never updated. When Epic changes the Verse spec, those tools silently produce wrong code.
UEFN Toolbelt is the only Verse code generator backed by the **live Verse language spec** — [verselang/book](https://github.com/verselang/book), a 27,000-line authoritative reference maintained by the Verse language team. Before generating any code, Claude queries the spec for exact syntax, effect specifiers, and API signatures. The generated code is spec-accurate, not template-accurate.
**How it works in practice:**
When you ask Claude to generate Verse code via the MCP bridge, it:
1. Calls `verse_book_search("suspends")` or `verse_book_chapter("concurrency")` to pull the relevant spec section
2. Reads the authoritative syntax and examples directly
3. Generates code that matches the current spec — not a six-month-old template
**Verse book MCP tools:**
| Tool | Description |
|---|---|
| `verse_book_search` | Search all 18 spec chapters by keyword — returns matching sections with context |
| `verse_book_chapter` | Fetch a full chapter by topic (`concurrency`, `classes`, `failure`, `effects`, etc.) |
| `verse_book_update` | `git pull` the latest spec from upstream — one call keeps you current |
**Keeping the spec current — one command, always:**
When Epic ships a Verse update, run either of these to pull the latest spec instantly:
```python
# Option 1: from inside Claude Code (MCP tool — no terminal needed)
verse_book_update()
# Option 2: from a terminal
git -C verse-book pull
```
That's it. Next time Claude generates Verse code it's working from the updated spec. No reinstall, no version bump, no waiting for a tool maintainer to push an update — you pull directly from the Verse language team's own repo.
**`verse_gen_custom` — Claude writes your device:**
```python
# Claude consults the spec, then writes your code to disk
tb.run("verse_gen_custom",
filename="my_patrol_ai.verse",
code="...", # Claude fills this from spec-verified generation
description="Patrol AI using spline waypoints and concurrency race block")
```
**Spec coverage:** expressions, primitives, containers, operators, mutability, functions, control flow, failure, structs, enums, classes, interfaces, types, access modifiers, effects, concurrency, live variables, modules, persistable state, language evolution — all 18 chapters.
---
## CLAUDE.md — AI-Native Repository (First of Its Kind)
Most GitHub repos require hours of reading before an AI can contribute meaningfully.
UEFN Toolbelt is the first UEFN project — and one of the first on GitHub at any scale —
to ship full AI-native onboarding as a core feature.
**How it works:** Claude Code automatically loads `CLAUDE.md` whenever you open the project directory. The file contains everything Claude needs to contribute correctly on the first try:
- Every tool name, category, parameter, and usage example across all 358 tools
- The exact nuclear reload command and when it's safe vs. unsafe (Quirk #26)
- V2 device property wall — what Python can and cannot configure (Quirk #19)
- UEFN path mounting quirks — why `/Game/` breaks and what to use instead (Quirk #23)
- Commit format, testing requirements, and the two-phase validation workflow
- MCP bridge command reference for AI → UEFN live control
- The complete 6-phase industrial pipeline for autonomous game building
**What this means in practice:**
```bash
git clone https://github.com/undergroundrap/UEFN-TOOLBELT
cd UEFN-TOOLBELT
claude
# → Claude knows all 358 tools, all UEFN quirks, and the full test workflow.
# → "Add a tool that does X" works on the first try. No session history needed.
```
Any contributor — including external contributors who've never seen this codebase — opens it with Claude Code and immediately has the same working knowledge as someone who's built it from scratch. The onboarding system is the repo.
> **No other UEFN tool on GitHub has this.** Most repos have a README. This repo has a machine-readable contributor brain that upgrades every AI session that opens it.
See [CONTRIBUTING.md](CONTRIBUTING.md) for the full 5-step contributor loop with Claude.
---
## Why This Is the Best UEFN Python Tool
UEFN Toolbelt is not just a collection of scripts; it is a **secure platform** for the UEFN community. It is the definitive way to **add, test, update, verify, and document** UEFN Python tools:
- **Adding & Updating**: Drop any Python script into the `Custom_Plugins` folder and it instantly acquires a beautiful UI card in the PySide6 Dashboard with its author, version, and external documentation links.
- **Security Verification**: Every plugin passes through a 4-Gate strict security audit (Size constraints, AST malicious import scanning, Namespace protection, and SHA-256 hashing) before it can ever execute. You can see the health of every script live in the UI.
- **Testing**: A programmatic `integration_test.py` validates that your tools work perfectly against the live UEFN C++ API, giving developers total confidence after engine updates.
| Feature | UEFN Toolbelt | Everyone else |
|---|---|---|
| Ecosystem Moat | **Rich Plugin Hub, automatic UI generation** | Scattered, undocumented gists |
| Security Model | **4-Gate Audit (AST scanning, SHA-256)** | Blind execution of untrusted code |
| Verification | **Automated Integration Test Suite** | Manual testing only |
| Tool count | **358 tools, 76 modules** | Single-purpose scripts |
| AI integration | **Full MCP bridge + model-agnostic HTTP client + tool manifest** | None |
| Local model support | **LM Studio, Ollama, any HTTP agent** | None |
| Verse code gen | **Live spec-backed (27K line reference)** | Static templates |
| Dashboard | **Sidebar nav, Plugin Hub, 13 pages, dark theme** | Blueprint widgets or none |
| Architecture | **Registry + decorators, undo-safe, extensible** | Monolithic scripts |
| AI onboarding | **CLAUDE.md auto-loaded — any contributor + Claude Code = full context instantly** | None |
| Day-1 release | **Shipped UEFN 40.00 launch day (March 2026)** | |
---
## Testing & Validation
The UEFN Toolbelt includes a professional-grade testing suite to ensure stability across UEFN updates.
### 1. Smoke Test (Healthy Schema Check)
Verifies all 76 modules are loaded, all 358 tool schemas are valid (descriptions, tags, `**kwargs` compliance), and UEFN API access is healthy.
```python
import UEFN_Toolbelt as tb
tb.run("toolbelt_smoke_test")
```
### 2. Integration Test (Context-Aware Verification)
A sophisticated fixture-based test that spawns actors, programmatically selects them in the viewport, runs tools, and verifies the results (e.g., material changes, transforms, JSON exports).
```python
import UEFN_Toolbelt as tb
tb.run("toolbelt_integration_test")
```
*Validated areas: Materials, Bulk Ops, Prop Patterns, Level Snapshots, API Crawler, Asset Tagger, Verse Helpers, and Screenshots.*
---
## Development Workflow
### Hot-Reloading (No UEFN Restart Required)
Use the Nuclear Reload command — it completely wipes the Toolbelt from Python's module cache and re-imports everything fresh. See the **Quick-Reference Command Table** above for the full set of variants (dashboard, graph window, tests, etc.).
```python
# Standard — reload everything and open the dashboard
import sys; [sys.modules.pop(k) for k in list(sys.modules) if "UEFN_Toolbelt" in k]; import UEFN_Toolbelt as tb; tb.register_all_tools(); tb.launch_qt()
```
> [!TIP]
> `deploy.bat` prints the right command for your current task after each deploy — no memorisation needed.
---
| Open source | **AGPL-3.0, full source** | Closed or single files |
Built for the 2026 UEFN Python wave. First. Most complete. Spec-accurate.
---
## Documentation
| File | What it covers |
|:---|:---|
| [docs/PIPELINE.md](docs/PIPELINE.md) | **The Industrialization Pipeline** — the master document. Every phase (0–6), every tool, current status, Claude's full execution script, the recursive build+fix loop, and the one remaining human step |
| [docs/AI_AUTONOMY.md](docs/AI_AUTONOMY.md) | **AI autonomy loop** — complete live run walkthrough, generated Verse file, how `world_state_export` + `verse_write_file` work, hard limits, roadmap to full headless builds |
| [docs/SCHEMA_DEEP_DIVE.md](docs/SCHEMA_DEEP_DIVE.md) | **Forensic schema analysis** — type taxonomy, inheritance evidence, enum full tables, network architecture, collision mask system, naming conventions, read-only vs writable |
| [docs/DEVICE_API_MAP.md](docs/DEVICE_API_MAP.md) | All 14 C++ actor classes — property tables, enums, automation patterns, access cheatsheet |
| [docs/FORTNITE_DEVICES.md](docs/FORTNITE_DEVICES.md) | Fortnite Creative devices — Trigger, Score Manager, Teleporter, Guard Spawner, channel wiring |
| [docs/SCHEMA_EXPLORER.md](docs/SCHEMA_EXPLORER.md) | How the schema system works — three layers, tool manifest, AI integration |
| [docs/UEFN_QUIRKS.md](docs/UEFN_QUIRKS.md) | 17 non-obvious UEFN Python behaviors — main thread lock, MCP return loop, ghost files |
| [docs/ui_style_guide.md](docs/ui_style_guide.md) | **UI Style Guide** — color palette, `ToolbeltWindow` API, widget recipes, canvas theming, AI agent rules. Read before writing any PySide6 UI. |
| [docs/plugin_dev_guide.md](docs/plugin_dev_guide.md) | Custom plugin development — skeleton, return contract, security model, UI style requirements |
| [docs/uefn_python_capabilities.md](docs/uefn_python_capabilities.md) | Full UEFN Python API surface — what's scriptable, what's read-only, what's stripped |
| [docs/CHANGELOG.md](docs/CHANGELOG.md) | Full version history — every feature, fix, and refactor by release |
---
## Patch Notes
### Unreleased — qFoldIT: Real-World Trademark Watchlists, UAG Scene Pipeline, Experiment Records
- **Real-world trademark watchlist extensions:** `qfoldit/compliance/watchlists/scientific_equipment_watchlist.json` and `vehicle_watchlist.json` — 22 lab/analytical-equipment brands (Thermo Fisher, Agilent, Danaher sub-brands, Bruker, Illumina, ...) and ~35 vehicle/heavy-equipment brands (Sherp, Kamaz, Caterpillar, Scania, Tesla, ...). English/Latin-script only by design (no native-script text anywhere in the files); several generic-word brand names shipped `enabled: false` to avoid false-positive noise. Loaded via the new `qfoldit/compliance/watchlist_loader.py` and `TrustRuntime.with_extended_watchlist()` — **the live `mcp_server.py` server now uses this by default**, not just in tests.
- **Fixed a real bug found while adding these:** `TrustRuntime._normalize()` was ASCII-only (`[^a-z0-9]`), which silently reduced any non-Latin (e.g. Cyrillic) watchlist term to an empty string — and an empty string matches everything as a substring. Fixed to be Unicode-aware; regression-tested.
- **UAG scene pipeline:** the `game-designer` Claude Skill (produces engine-neutral Universal Assembly Graphs) and the new `unreal-world-builder` Claude Skill (the only engine adapter in this repo — realizes a UAG via real, grep-verified `run_toolbelt_tool` calls, and explicitly reports node types it can't yet realize instead of skipping them) now live under `.claude/skills/`. `qfoldit/science/uag_bridge.py` + the new `qfoldit_gamedesign_to_uag_seed` tool deterministically bridge a game design document into a starter UAG.
- **Experiment Record system:** `qfoldit/science/experiment_record.py` + `qfoldit_build_experiment_record` ties a science result, its `TrustRuntime` licensing decisions, and its game-design/UAG context into one sha256-identified, citation-grounded, publication-checklist-carrying record — see [qFoldIT Science, Gamification & Experiment Pipeline](#qfoldit-science-gamification--experiment-pipeline). Records persist to `qfoldit/logs/experiment_records.log.jsonl` (`qfoldit_list_experiment_records` reads them back).
- **Closed the scene-object-licensing gap:** new `TrustRuntime.decisions_since()`/`now_ts()` read the runtime's own audit log back out; new `qfoldit_scene_build_start`/`qfoldit_collect_scene_licensing` tools plus `qfoldit_build_experiment_record`'s `auto_collect_licensing_since` argument mean an experiment record can now prove every game object placed while building a scene was checked, not only the science-tool calls that fed it.
- **34 new tests** (`tests/test_qfoldit_watchlist_extensions.py` — 8, `tests/test_qfoldit_uag_bridge.py` — 9, `tests/test_qfoldit_experiment_record.py` — 9, `tests/test_qfoldit_scene_licensing_collection.py` — 8). **Total: 85/85 qFoldIT tests passing.**
### Unreleased — qFoldIT Moved Into Its Own Folder + Real Science Pipeline Wired In
- **Reorganized:** everything qFoldIT-related moved from loose files at the repo root into [`qfoldit/`](qfoldit/README.md) — `compliance/`, `science/` (with a new `pipelines/` subpackage), `monetization/`, each with its data file(s) next to the code that reads them. Runtime logs moved to `qfoldit/logs/`. All default file paths now resolve relative to the module's own location, not the process's CWD — matters because `mcp_server.py` may launch from inside the UEFN editor's own working directory.
- **Real science pipeline added:** `qfoldit/science/pipelines/quantum_runner.py` (classical quantum-walk-inspired folding simulation, real NeRF backbone reconstruction, plus an import-guarded real CVaR-VQE wrapper) and `qfoldit/science/gamedesign.py` (deterministic game-design-document generator) — both pure-stdlib, ported from the original qFoldIT/Protein-Design-MCP repo. Exposed as three new MCP tools: `qfoldit_quantum_walk_fold`, `qfoldit_quantum_vqe_fold`, `qfoldit_generate_game_design`.
- **Found and documented a real algorithmic limitation, not hidden:** `simulate_quantum_walk_fold`'s single-dihedral-angle Metropolis moves can leave chains of ~14+ residues kinetically trapped in a self-clashed conformation — confirmed by a dedicated test (`test_known_limitation_longer_chains_can_stay_kinetically_trapped`), documented in the function's own docstring.
- **20 new tests** (`tests/test_qfoldit_quantum_runner.py` — 12, `tests/test_qfoldit_gamedesign.py` — 8, including a real end-to-end run through both modules together). **Total: 51/51 passing.**
### Unreleased — qFoldIT Trust & Compliance Layer Merged
- **IP compliance gate** (`qfoldit_trust_runtime.py`): `run_toolbelt_tool` and `execute_python` in `mcp_server.py` now pass through a default-deny watchlist/manifest check before reaching the live editor. Covers LEGO, TMNT, The Walking Dead Universe, Squid Game, KPop Demon Hunters, Star Wars, and flags anything else third-party by default — see [qFoldIT Trust & Compliance Layer](#qfoldit-trust--compliance-layer).
- **Real asset provenance verification**: `TrustRuntime` now accepts an `asset_metadata_fn` hook wired to the live editor's own Asset Registry (via the existing `get_asset_info` IPC command — no network call). Path claims in a tool call are resolved through the engine instead of trusted as text; a mismatch (text says Epic-native, engine disagrees) is blocked as spoofing rather than silently allowed. `license_manifest.json` entries gained `content_plugin_ids` for real plugin-level matching per brand.
- **Boltz commission pricing** (`qfoldit_monetization_registry.py` + `boltz_pricing.json`): `qfoldit_evaluate_commission` now accepts `sequence_length`/`num_samples` and returns a local, offline compute-time + cost estimate for Boltz-backed commissions. No network lookup — consistent with this repo's zero-outbound-calls guarantee — and no invented dollar figures: `estimated_cost_usd` stays `null` until `rate_usd_per_gpu_second` is filled in from your real Modal billing.
- **Dashboard visualization**: new `qfoldit_trust_dashboard` tool (category: Dashboard) — read-only PySide6 panel summarizing the compliance audit log (allowed/blocked, provenance method breakdown), licensed brands with royalty/review status, and the commission ledger (accepted/blocked, priced vs. unpriced totals). Registered the standard 3-step way in `tools/__init__.py`, no changes to `dashboard_pyside6.py` required.
- **Science MCP Registry** (`qfoldit_science_mcp_registry.py`): separate `verified`/`connected`/`best_effort`/`reference_only` gate for which scientific MCP servers auto-connect vs. require explicit opt-in.
- **31 tests** across `tests/test_qfoldit_trust_runtime.py` (16), `tests/test_qfoldit_science_mcp_registry.py` (7), `tests/test_qfoldit_monetization_registry.py` (8). Run with plain `python3`, no UEFN editor needed.
### v1.5.3 — March 2026 (Audit Fixes + MCP Auto-Start)
- **Version string corrected**: `__version__` bumped from stale `1.2.0` to `1.5.3`. Dashboard About tab now shows `v1.5.3 · 247 tools`.
- **MCP listener auto-starts** when the dashboard opens — no more manual `tb.run("mcp_start")`. If Claude Code is already configured, it connects automatically on first dashboard open. Silent if already running; never blocks the UI if MCP fails.
- **Smoke test threshold corrected**: `MIN_TOOL_COUNT` updated from stale `155` to `171`. Smoke test now catches regressions at the real tool count.
- **Safe tool execution expanded**: 23 no-argument tools now verified at smoke-test time (was 9). Coverage: `theme_list`, `theme_get`, `config_list`, `config_get`, `verse_graph_scan`, `api_search`, `spline_measure`, and more.
- **Coverage report tool upgraded** (`list_untested.py`): Fixed repo root path resolution, improved string literal detection, categorized output, CI-friendly exit codes (0 = all covered, 1 = gaps). Run `python Content/Python/UEFN_Toolbelt/list_untested.py` any time to see 78% coverage with a grouped gap report.
### v1.5.2 — March 2026 (Live Theme Switcher + Appearance Tab)
- **6 built-in themes** (`toolbelt_dark`, `midnight`, `ocean`, `nord`, `forest`, `daylight`): Switch via the new **Appearance** tab in the dashboard, `tb.run("theme_set", name="ocean")`, or MCP. Changes apply live to every open window instantly — no restart needed.
- **Appearance tab** added to the dashboard sidebar with visual swatch buttons. Each swatch is styled with its own theme colors so you see exactly what you're choosing. Active theme shown with ✓ marker and white border.
- **Theme tools** (`theme_list`, `theme_set`, `theme_get`): 3 new registered tools. `theme_set` persists the choice to `config.json` and notifies all subscribers. `theme_get` returns the full palette dict so AI agents can introspect current colors.
- **Subscriber system** in `core/theme.py`: `subscribe(fn)` / `unsubscribe(fn)` — any window can register for live theme changes. `_ToolbeltDashboard` and all `ToolbeltWindow` subclasses subscribe automatically on open and unsubscribe on close. Dead Qt object callbacks cleaned up automatically via `RuntimeError` detection.
- **Tool count: 171** (up from 168).
### v1.5.1 — March 2026 (Theme System + ToolbeltWindow)
**Why this matters:** As the platform grows — more tool windows, community plugins, AI-generated features — the only way to keep everything looking professional is to make consistency *structural*, not just documented. This release does that.
- **`core/theme.py`** (new): `PALETTE` dict is the single source of truth for every color in the platform. `QSS` is built dynamically from `PALETTE` — edit one hex value and the dashboard, every tool window, and every plugin updates automatically on next reload. No more hunting across files.
- **`core/base_window.py`** (new): `ToolbeltWindow(QMainWindow)` base class. Subclass instead of `QMainWindow` directly to get: theme applied automatically, Slate tick driver via `show_in_uefn()` (required in UEFN or windows are invisible), `self.P` palette dict, `self.hex(token)`, and factory helpers for every standard widget type (`make_topbar`, `make_btn`, `make_label`, `make_divider`, `make_text_area`, `make_hbar`, `set_hbar_value`, `make_scroll_panel`). What used to be ~45 lines of boilerplate per window is now 3 lines.
- **`dashboard_pyside6.py`**: `_QSS` now sourced from `core/theme.py` instead of defined inline. Backward-compatible.
- **`verse_device_graph.py`**: Refactored to use `ToolbeltWindow`. ~40 lines of manual boilerplate removed.
- **`docs/ui_style_guide.md`**: Fully rewritten — documents the new architecture, full `ToolbeltWindow` API, canvas/QGraphicsScene theming, semantic color table, and a dedicated AI agent rules section.
- **`docs/CHANGELOG.md`** (new): Full version history.
- **Hot-reload command table** added to README and `deploy.bat` — one command per dev task, always at hand.
### v1.5 — March 2026 (Verse Device Graph + Config Persistence)
- **Verse Device Graph** (`verse_graph_open`, `verse_graph_scan`, `verse_graph_export`): The most powerful tool in the Toolbelt for understanding a UEFN level's logic architecture. Interactive node graph of every Creative/Verse device — nodes grouped in Blueprint-style category columns, edges colored by type (`@editable` red, `.Subscribe` green, `.call` blue). Architecture Health Score (0–100) with Union-Find cluster detection. **Minimap** overlay (colored dots + blue viewport rect, click/drag to navigate). **Category filter** dropdown to isolate a device family. **Focus** button to jump to any selected node. Comment/note boxes survive every re-scan. Hover highlight dims unrelated nodes. ● Live mode polls every 4 s without disturbing node positions. Write-back renames actors and moves folders from the side panel. Gen Wiring generates a ready-to-compile `creative_device` stub. **?** Help dialog covers full purpose, workflow, badge guide, and tips. Fully MCP-callable via `verse_graph_scan`. *(Inspired by [ImmatureGamer's uefn-device-graph](https://github.com/ImmatureGamer/uefn-device-graph) — independent PySide6 rewrite integrated into the Toolbelt stack.)*
- **Persistent Config System** (`config_list`, `config_get`, `config_set`, `config_reset`): 12 configurable values persisted at `Saved/UEFN_Toolbelt/config.json` — survives `install.py` updates. Tools read from config instead of hardcoded defaults. `verse.project_path` lets you set your Verse root once and never type it again.
- **Tool count: 171** (up from 165). 103/103 integration tests passing.
### v1.4 — March 2026 (Deep Schema Documentation)
- **`docs/DEVICE_API_MAP.md` rewrite**: Complete property reference for all 14 C++ actor classes in the reference schema. Every readable property with type, default value, and automation notes. Enum value tables (`FortBuildingType`, `FortResourceType`, `NetDormancy`, etc.). The 19 restricted properties explained. Property access cheatsheet (`get_editor_property` vs `getattr` vs component access).
- **`docs/FORTNITE_DEVICES.md`**: New document covering every common Fortnite Creative device (Trigger, Team Settings, Score Manager, Guard Spawner, Teleporter, Item Spawner, Capture Area, Countdown Timer). The Python↔Verse bridge problem documented with two working access paths. Channel system (1–255) explained with standard channel conventions. Full `export → inspect → automate` workflow. Verse code generation quick reference.
- **`docs/SCHEMA_EXPLORER.md`** updated: Three-schema-layer model (tool manifest + C++ schema + Verse schema). Schema files reference table. Phase 21 complete AI return loop section.
- **`docs/uefn_python_capabilities.md`** updated: Tool count corrected to 161.
### v1.3 — March 2026 (Phase 21: Complete AI Return Loop)
- **100% Structured Dict Returns**: Every single `@register_tool` function across all 23 tool modules now returns `{"status": "ok"/"error", ...data...}`. This is the full completion of the MCP return contract — AI agents calling any tool via the bridge receive a machine-readable result. Zero `None` returns, zero bare primitives, zero unreal objects remain.
- **`describe_tool` MCP Command**: New MCP bridge command that returns a single tool's full manifest entry (`name`, `description`, `parameters`, `tags`, `category`) without loading the entire `tool_manifest.json`. AI agents can query parameter contracts per-tool on demand.
- **Files updated**: `spline_prop_placer`, `text_painter`, `smart_organizer`, `localization_tools`, `foliage_converter`, `sequencer_tools`, `lighting_mastery`, `verse_schema`, `system_build` — all 15+ remaining `-> None` and primitive-return tools converted.
- **Internal callers updated**: `run_measure_travel_time`, `run_enforce_conventions (dry_run path)` — callers that expected float/None return values updated to handle the new dict returns.
### v1.2 — March 2026 (Phase 20: AI-Agent Readiness)
- **358 tools** across 55 categories
- **Tool Manifest Export** (`plugin_export_manifest`): Writes `Saved/UEFN_Toolbelt/tool_manifest.json` — a machine-readable index of every registered tool with its full parameter signature (name, type, required/optional, default). Any AI agent or automation script can load this file and know how to call every tool without reading source code. This is the key artifact for full AI-driven UEFN workflows.
- **Structured Returns Everywhere (Phase 21 complete)**: All 358 tools return `{"status": "ok"/"error", ...}` dicts. Zero `None` returns remain. MCP callers (Claude Code, `client.py`, scripts) can read every result programmatically — no log parsing required.
- **Schema-Driven Property Discovery** (`schema_utils.discover_properties`): `verse_device_editor`'s property reader now queries the reference schema for each actor's class before falling back to a hardcoded list. It reads whatever properties the schema actually defines for that class, making it correct-by-construction rather than hardcoded.
- **Registry `to_manifest()`**: New method on `ToolRegistry` that introspects every function's signature via `inspect.signature()` — captures param names, type annotations, required/optional status, and defaults. Powers `plugin_export_manifest` and exposes the full tool catalog programmatically.
- **`schema_utils` expansion**: Added `list_classes()` (all schema class names) and `discover_properties(class_name)` (schema property dict for a class) — two new helper functions that replace hardcoded property lists with live schema lookups.
- **Full MCP Return Loop**: The complete chain is verified: tool returns dict → `registry.execute()` → `_serialize(result)` → JSON in MCP response → readable by Claude Code. Every structured return flows end-to-end.
### v1.1 — March 2026 (Phase 19: Simulation & Sequences)
- **160 tools** across 50 categories: Simulation, Sequencer, Verse Helpers, and more
- **Verse Simulation Proxies**: Generate Python counterparts for @editable Verse properties to test logic without full UEFN sessions.
- **Named Auto-Link Breakthrough**: Robust fuzzy resolution for `VerseDevice_C` actors—auto-links viewport labels to Verse schema.
- **Sequencer Automation**: One-click "Actor to Spline" paths and batch keyframe management.
- **Dashboard v2 Refactor**: Migrated ALL 24 tabs to the modern `builder(R)` pattern; fixed critical `NameError: _title` bug.
- **Deep Diagnostics**: New `debug_dump_verse_actor` and `debug_audit_verse_assets` tools for troubleshooting hidden API links.
### v1.0 — March 2026 (Initial Release)
- **250 tools** across 13 categories: Materials, Procedural, Bulk Ops, Text, Assets, Verse, Project, Screenshot, Tags, MCP, API Explorer, Utilities, and more
- **PySide6 dashboard** — dark-themed floating window with sidebar nav, search across all tools, and per-category pages
- **Tool Registry** — `@register_tool` decorator system, execute-by-name, tag/category search
- **MCP bridge** — full two-process architecture letting Claude Code control UEFN over HTTP
- **Verse spec integration** — live `verselang/book` backed Verse code generation
- **Smoke test** — 6-layer health check (Python env, UEFN API, tool registry, MCP bridge, PySide6, Verse spec)
- **UEFN 40.00 compatibility** — tested against real UEFN API, fixed `TextRenderHorizontalAlignment`, `find_or_add_section`, `register_menu` keyword arg incompatibilities
- **About page** — in-dashboard credits, license info, and repo links
---
## Why the Theme System Matters
Most tools hard-code their colors. When a new window is built, the developer copy-pastes hex values and the UI slowly drifts — different shades of dark, slightly different borders, inconsistent accent colors. Community plugins make it worse because plugin authors have no reference to match.
The Toolbelt theme system solves this structurally, not just with documentation:
- **`core/theme.py` is the single source of truth.** `PALETTE` is a live dict. Every color in the platform — dashboard, tool windows, device graph, community plugins — reads from it. Change one value and every window updates on the next reload.
- **`set_theme()` updates everything live.** It mutates `PALETTE` in-place, rebuilds the QSS stylesheet, and notifies every subscribed window via a callback list. No restart needed. This is exactly how Discord, VS Code, and Obsidian implement live theme switching.
- **`ToolbeltWindow` enforces the standard automatically.** Subclassing `ToolbeltWindow` instead of `QMainWindow` means your window gets the theme applied, subscribes to future changes, and unsubscribes cleanly on close — for free. Plugin authors get platform-consistent UI without knowing how the theme system works.
- **Community plugins inherit theming.** A plugin that subclasses `ToolbeltWindow` will switch themes with the rest of the platform automatically. The platform grows without visual fragmentation.
- **6 built-in themes** (`toolbelt_dark`, `midnight`, `ocean`, `nord`, `forest`, `daylight`) — switchable from the Appearance tab, `tb.run("theme_set", name="ocean")`, or MCP. Persists across restarts via `config.json`.
This is what separates a collection of scripts from a platform.
---
## Attributions
The UEFN Toolbelt is built on the shoulders of community pioneers who proved these patterns work:
| Contributor | Project | What they pioneered |
|---|---|---|
| **[Kirch](https://github.com/KirChuvakov/uefn-mcp-server)** ([@KirchCreator](https://x.com/KirchCreator)) | [uefn-mcp-server](https://github.com/KirChuvakov/uefn-mcp-server) | MCP bridge for UEFN — the queue + Slate tick pattern for calling `unreal.*` from external processes. The Toolbelt MCP bridge is an independent rewrite extending this architecture. |
| **[ImmatureGamer](https://github.com/ImmatureGamer/uefn-device-graph)** ([@ImmatureGamer](https://x.com/ImmatureGamer)) | [uefn-device-graph](https://github.com/ImmatureGamer/uefn-device-graph) | Verse device graph concept — first tkinter implementation of a UEFN node graph tool. The Toolbelt device graph is an independent PySide6 rewrite built into the Toolbelt stack. |
Both tools are independently developed and go significantly beyond their inspirations, but these creators deserve credit for proving the concepts first. Go follow them.
---
## Contributing
Contributions are welcome and encouraged. This project follows a simple rule: **keep it useful for real UEFN creators.**
> **Testing your tools:** Before contributing, please read [TOOL_STATUS.md](TOOL_STATUS.md) to understand which tools are covered by automated tests and which require manual verification.
### How to contribute
1. **Fork** the repo and create a branch: `git checkout -b feature/my-tool`
2. **Follow the tool structure** — see [Adding a New Tool](#adding-a-new-tool) for the exact pattern
3. **Test in a live UEFN editor — this is mandatory.** Syntax checks don't catch UEFN runtime failures. Run the hard-refresh bundle in the UEFN Python console and confirm your tool works:
```python
import sys; [sys.modules.pop(k) for k in list(sys.modules) if "UEFN_Toolbelt" in k]; import UEFN_Toolbelt as tb; tb.register_all_tools(); tb.run("your_tool_name")
```
Then run the smoke test to confirm no regressions: `tb.run("toolbelt_smoke_test")`
4. **Open a Pull Request** with a clear description of what the tool does and why it belongs here
### Guidelines
- Every tool must be wrapped in `ScopedEditorTransaction` if it modifies the level — one Ctrl+Z must undo it
- No network calls. Tools that need external data should require the user to provide it as a parameter
- No new dependencies beyond `unreal` (built-in) and `PySide6` (already required for the dashboard)
- Tool names use `snake_case`. Category names use `Title Case`. Both must be unique
- Add your tool to the smoke test if it has a meaningful self-test
- If you fix a UEFN API incompatibility, document what the correct API is in a comment — the API is underdocumented and your discovery helps everyone
### Code style
- Follow the existing patterns in `tools/` — no need for docstrings on every line, but the `@register_tool` description must be clear enough for the search box
- Python 3.11 compatible — no f-string backslashes, no 3.12+ syntax
- All paths use `unreal.Paths` or `os.path` — no hardcoded Windows separators
### Attribution
By contributing, you agree that your code may be distributed under the AGPL-3.0 license. Your name will be credited in the patch notes for the release that ships your contribution.
---
## Tool Requests
Have an idea for a tool that should be in the Toolbelt? Open a GitHub Issue with the label **`tool-request`**.
**Good tool request format:**
```
Title: [Tool Request] Bulk set collision preset on selected actors
What it does:
Sets the collision preset (e.g. BlockAll, OverlapAll, NoCollision) on all
selected static mesh actors in one click.
Why it's useful:
Setting collision on 50 props one at a time in Details is extremely slow.
This is one of the most common repeated tasks in UEFN prop placement.
Category: Bulk Ops
Tags: collision, bulk, static mesh
```
The more specific you are about the exact UEFN operation and why it saves time, the more likely it is to ship.
> **Note:** Tool requests from contributors who have already opened a PR get prioritized.
You can also request tools directly from inside the dashboard — click **About → Open an Issue on GitHub**.
---
## License
**GNU Affero General Public License v3.0 (AGPL-3.0) with Visible Attribution Requirement**
Copyright © 2026 **Ocean Bennett**
This software is free and open source. You may use, fork, and modify it under AGPL-3.0. Derivative works — forks, plugins, tools built on top of this codebase — must also be open source and must include visible credit:
> "Built on UEFN Toolbelt by Ocean Bennett (https://github.com/undergroundrap/UEFN-TOOLBELT)"
**Commercial use:** Integrating this toolbelt into a closed-source or monetized product requires a separate commercial license. Contact Ocean Bennett to arrange terms.
Full license: see [`LICENSE`](LICENSE) — or read it in the dashboard under **About → License**.
### qFoldIT-contributed components
The compliance/science/gamification/experiment-record layer under [`qfoldit/`](qfoldit/README.md) and
[`.claude/skills/`](.claude/skills/game-designer) — see
[qFoldIT Trust & Compliance Layer](#qfoldit-trust--compliance-layer) and
[qFoldIT Science, Gamification & Experiment Pipeline](#qfoldit-science-gamification--experiment-pipeline)
above — was merged from the qFoldIT team into this repository and ships as part of this same codebase,
under the same AGPL-3.0 terms above, unless a `LICENSE`/`NOTICE` file inside `qfoldit/` states otherwise.
No such file exists in `qfoldit/` at present — if the qFoldIT team requires separate terms for their
components, that should be added as an explicit `qfoldit/LICENSE` or `qfoldit/NOTICE` file rather than
assumed.
---
*Built by Ocean Bennett — [@undergroundrap](https://github.com/undergroundrap) — for the 2026 UEFN Python wave.*
*qFoldIT extension merged from the qFoldIT team — see [qFoldIT Trust & Compliance Layer](#qfoldit-trust--compliance-layer).*
This server cannot be deployed
Maintenance
ActivityMaintained
ResponsivenessUnresponsive