NEOSTACK_AI.md

pipelines/NEOSTACK_AI.md

NeoStack AI — what it actually is, mid-2026

Status: research brief, not canon. Written to resolve docs/pipeline_review/STACK_FACTS_QUESTIONS.md

Q1 ("NeoStack confirmation"). Research date 2026-07-15, web-verified (training-data cutoff was Jan

2026; UE 5.8 itself shipped June 17 2026, after cutoff, so this whole space had to be checked live).

Every claim is tagged VERIFIED (checked against a primary/reputable source, URL inline) or

INFERRED (my synthesis bridging a gap sources didn't state directly). Where sources actively

conflicted, I say so instead of picking one silently.

0. TL;DR verdict — read this first

**"Neostack AI" is real, unambiguous, and is the same product the contract docs already call

"NeoStack."** Q1's recommendation ("confirm-same") is CORRECT — apply it. VERIFIED:

1. The product: NeoStack AI, made by Betide Studio, Inc. (a real, established Unreal Engine

plugin developer — founded 2021, Lucknow India, founder Devesh Mishra, 40,000+ developers using

their catalog — Tracxn company profile).

Sold both direct (neostack.dev) and on Epic's Fab marketplace under the

listing title "Agent Integration Kit (NeoStack AI)"

(fab.com listing). It is an

Unreal Editor plugin that lets an external AI coding agent — **Claude Code by name, first in every

list** — drive the editor directly: author/edit Blueprints, Materials, Niagara, Sequencer

cinematics, Behavior/State Trees, Widget Blueprints, Data Tables, PCG graphs, IK Rigs, with live

edits (not suggestions) and full undo support. This is functionally *exactly* what

T1_Build_Pipeline_Contracts §1.14 describes as "the Claude tool surface inside the UE5 editor" —

§1.14's own words. No other product this thorough matches that description this closely.

2. It is not the same thing as Epic's own native UE 5.8 "MCP Server (Experimental)" plugin —

a separate, first-party, free, bundled-with-the-engine feature that also showed up mid-2026 and

is easy to conflate with NeoStack because both are "an MCP thing that lets Claude drive the

editor." They are different products from different companies with very different maturity and

coverage. §8 below is a full compare — this distinction is the single most load-bearing finding

in this brief, because this repo's own docs currently blur the two (§9).

3. Real risk found, not previously known to this repo: every corroborated source puts NeoStack's

confirmed Unreal Engine support at 5.5/5.6/5.7 (docs.neostack.dev says a 5.4 floor). **No

source found confirms UE 5.8 support**, and UE 5.8 (the project's pinned engine target) only

shipped 2026-06-17 — about four weeks before this research. The most recent version data point

found (Fab listing metadata, v1.0.74, dated 2026-04-24) predates 5.8 entirely. This is a genuine

open compatibility question, not a paperwork gap — flag it, don't assume it resolves itself.

4. NeoStack has no fixed, versioned "input contract" API doc a pipeline can code against in the

sense this repo's Translation Doc concept wants. Its documented interface is conversational

(natural-language prompts into an "Agent Chat" panel) mediated by one of two connection modes —

ACP (Agent Client Protocol, for driving it from an existing agent CLI like Claude Code) or

its own MCP server (for MCP-compatible clients). The only genuinely stable, versioned,

public specs underneath either mode are the protocols themselves (ACP, MCP) — not NeoStack's

internal tool names, which the existing contract doc already correctly hedges as "lock at NeoStack

plugin documentation, not at this contract" (§1.14, verbatim). That hedge was the right call; keep

it.

5. Pricing is solo-dev-friendly and commercially unrestricted: core plugin is a one-time purchase

(~$110), bring-your-own AI subscription (Claude Code, Codex, Gemini CLI, Cursor, OpenRouter, or

direct API keys — "no credits burned" by NeoStack itself). An optional $20/mo "NeoStack Cloud"

tier (hosted IDE, version control, CI/CD, crash analytics) is layered on top and is skippable for a

solo dev not needing hosted build infrastructure. Fab's Standard License gates only on gross

revenue ($100k/12mo threshold, Personal vs Professional tier) with identical rights either way — no

special carve-out against a solo commercial game.

Bottom line recommendation: keep NeoStack as the pinned Claude↔UE5-editor tool surface (Q1:

confirm-same, apply as recommended) — but (a) verify actual UE 5.8 compatibility directly before the

Phase 6 vertical slice build, since no public source confirms it yet, (b) do not let this repo's

"UE 5.8 MCP Server plugin" phrasing (used in MILESTONE_BUILD_PIPELINE.md) keep meaning two different

products depending on which doc you're reading, and (c) treat §1.14's richer tool-category claims

(World Partition, Cesium, Gaea, SRTM, MetaHuman-Creator-invocation, Save-state, Quest/Combat

authoring) as unverified against NeoStack's own public documentation — some are plausible

extensions of confirmed capabilities, several have no corroboration at all. Full breakdown in §9.

---

1. Identification — is this really the product, and is the name ambiguous?

**Name-collision check (requested by the brief): yes, "Neostack" collides with unrelated

companies, but none of them are remotely adjacent to game dev or AI agents**, so there is no real

ambiguity risk for this specific research question. Distinct, unrelated "Neostack"/"Neo Stack"

entities found: a South Korean IoT/smart-lighting mesh-connectivity company (founded 2018), a

Bangalore React/GraphQL/React-Native consultancy (founded 2018), a Tomsk, Russia IT-consulting firm,

and "NeoStacked" (a generic AI/cloud web-dev shop) — VERIFIED via

general web search results returning LinkedIn/Tracxn/CBInsights profiles

for each. None publish anything about Unreal Engine, game development, or Claude/agent tooling.

The product that matches the brief's description ("Claude/AI↔UE5-editor tool surface") is

unambiguously the Betide Studio one — every independent source (official site, Fab listing,

developer blog coverage, YouTube walkthroughs, Epic forums) converges on the same product, company,

and feature set. **Not a case of "genuinely cannot be found" or "ambiguous name" — this one is

clean.**

Also worth flagging: Betide Studio separately runs "NeoAI" at

neodocs.betide.studio, described as "a specialized AI suite

designed to power and accelerate Unreal Engine development," with an X/Twitter announcement from

Betide Studio calling it "our new AI assistant designed for Unreal Engine Blueprint developers...

fully agentic now — Blueprint generation, debugging, and real-time contextual help... built on a

custom model trained on a handcrafted dataset"

(Betide Studio on X). VERIFIED that NeoAI

exists as a distinct product page; NOT verified how NeoAI and NeoStack AI relate — predecessor

folded into NeoStack, sibling product, or same thing under two names. Neither product's own docs

cross-reference the other in what was fetched. This is a minor naming-confusion risk worth a

one-line disambiguation if this repo ever cites "NeoAI" — but every contract-doc reference in this

repo says "NeoStack," which is the product identified with high confidence above, so this doesn't

change the Q1 answer.

2. What it is, architecturally

NeoStack is an Unreal Editor plugin (Slate-based UI, native to the engine) that gives an AI

agent a structured tool surface over the project rather than being an agent itself. Betide Studio's

own framing: "NeoStack itself is not a pipeline — it is the tool layer that lets Claude operate as a

build agent" — which, notably, is *word-for-word the same framing* `T1_Build_Pipeline_Contracts

§1.14` already uses about NeoStack, another point of convergence supporting Q1's confirm-same call.

VERIFIED via betide.studio/neostack and

docs.neostack.dev.

Positioning has broadened over the product's life from a pure editor-bridge into an "AI-integrated

development suite" — the current marketing bundles the core plugin with an IDE, version control,

CI/CD, and crash analytics under one subscription umbrella (the Cloud tier, §6). The core

editor-bridge functionality (the part relevant to "drive the UE5 editor from an agent") is the

one-time-purchase plugin; the surrounding suite is optional.

3. Input contract — how you actually drive it

Primary interface: natural language, typed into an in-editor "Agent Chat" panel (Window / Tools

→ Agent Chat). Example from the docs: a user types something like "Create a Blueprint called

BP_HealthPickup that..." in plain English, and the plugin's own tool layer parses intent, calls the

right internal tools, and executes live in the editor. Supports "context attachments" — attaching a

specific Blueprint node or asset to a prompt — and does automatic project indexing for

context-awareness. VERIFIED, docs.neostack.dev.

Two ways an external agent pipeline can actually reach it:

1. ACP (Agent Client Protocol) — connects an already-running agent CLI directly to the editor.

Named-supported agents: Claude Code, Gemini CLI, Codex, Cursor, GitHub Copilot — "or any

ACP-compatible agent." In this mode, you run your normal agent CLI (e.g., claude in headless/

orchestration mode) and NeoStack is the editor-side surface presenting UE-specific tools to that

agent's tool-use loop. VERIFIED, exact quote from

betide.studio/neostack: "Connect AI coding agents — Claude

Code, Gemini CLI, Codex, Cursor, or any ACP-compatible agent — directly to the Unreal Editor,"

with a supported-agents list explicitly tagging each as "— via ACP."

2. MCP server mode — "or connect via the MCP server," i.e., NeoStack can also be reached by any

MCP-compatible client, not just ACP-native CLIs. VERIFIED, same source, same page.

Neither the ACP nor MCP acronym is spelled out on Betide's own marketing pages (I had to fetch the

page directly and quote it verbatim to confirm which protocols, since search-summarized results

dropped the acronyms) — worth knowing if anyone goes hunting in NeoStack's own docs expecting a

glossary.

What this means for an "input contract" a pipeline can be written against: there is no

NeoStack-specific structured command schema published as a stable third-party-facing API. The

contract surface that's actually stable is the protocol layer (ACP or MCP, both open,

versioned, publicly specified — ACP by Zed Industries, MCP by Anthropic), not NeoStack's internal

tool catalog, which Betide can and does change between releases (see version churn in §5). This

matches — and validates — the existing contract doc's own hedge: "Per-category tool inventory and

command syntax lock at NeoStack plugin documentation, not at this contract" (§1.14, verbatim

quote). **Recommendation: keep that hedge exactly as written; do not try to pin NeoStack's internal

tool names into a versioned contract doc — pin the protocol choice (ACP vs MCP) and the tool

*categories* required, which is what §1.14 already does.**

4. Capabilities — what it can actually do inside the editor

VERIFIED (marketing copy, corroborated across betide.studio, neostack.dev, and the Fab listing

title/description, plus a beta tester's firsthand account — see §5): **27+ built-in tools, 36+ asset

type readers, 150+ distinct operations** (Betide's own comparison-page numbers). Confirmed subsystem

coverage:

Not found in any NeoStack source (i.e., claimed nowhere in Betide's own marketing/docs, despite

being asked for explicitly): World Partition streaming/data-layer/HLOD configuration, Cesium Ion

tileset/georeference operations, Gaea import operations, SRTM heightmap import, MetaHuman Creator

brief invocation, C++ authoring/compilation. This gap matters — see §9, since the existing contract

doc attributes several of these to NeoStack by name.

5. Limits, maturity, and reliability reputation

Version/maturity timeline (reconstructed from multiple partially-conflicting sources — flagged

where they disagree):

• Search aggregation independently reported a Fab store beta launch of 2026-01-20, with **beta

v0.3.1 on 2026-02-09. Philip Conrod's developer blog, dated 2026-02-09**, is a firsthand

account of testing that same v0.3.1 beta and explicitly says stability is still a live concern

("It is still in Beta!"), that Betide "added almost 500+ additional checks... to reduce the number

of crashes," and that Betide Studio "currently recommends using Claude Code as it works best for

creating Blueprint logic while the other models produce less than desired results" — VERIFIED,

philipconrod.com.

• Fab listing metadata found via search snippet shows v1.0.58 on UE 5.7 and **v1.0.74 dated

2026-04-24** — i.e., by spring 2026 the plugin had graduated out of 0.x beta numbering into a 1.0.x

line. This is the most recent hard version data point found in this research; nothing newer was

located (see §0 point 3 — this predates UE 5.8 by two months).

• One fetched page (betide.studio/neostack) rendered a page-header artifact reading "February

2025|Betide Studio." This conflicts with everything above (a Jan 2026 Fab launch is inconsistent

with a Feb 2025 ship date, and Claude Code / multi-agent-CLI ACP interop as a mature ecosystem

didn't really exist in Feb 2025). **Flagging as likely a page-furniture artifact (e.g., a

copyright/blog-post date picked up by the fetch) rather than a real product-launch date** — do not

treat "February 2025" as reliable without independent confirmation.

Reliability signals: community-reported "100+ crash fixes shipped within days of launch," Discord

issue reports "resolved same-day," an acknowledged beta-era bug ("flickering effects"), and — the

most operationally important reliability fact — **output quality is bottlenecked by which LLM

drives it**: Betide's own recommendation and the beta tester's independent experience agree Claude

Code gives the best Blueprint-authoring results, with weaker/other models (used via OpenRouter's

400+ model catalog) producing "less than desired results." For this project (Claude-centric already)

this is favorable, not a blocker — but it means NeoStack's reliability reputation is really "reliable

when driven by Claude," not a model-agnostic guarantee.

Human-in-the-loop: NeoStack does not appear to enforce a pre-execution approval gate by design —

edits apply live, with full undo support as the stated safety net, and an "AI code review"

capability exists as a separate (Cloud-tier) feature rather than a mandatory checkpoint. INFERRED:

for an autonomous build loop, this means the safety boundary NeoStack itself provides is

"reversible," not "reviewed-before-applied" — any mandatory human-checkpoint policy has to be

imposed by the orchestrating pipeline (the Translation Doc / build-agent loop), not assumed from the

tool.

UE version support — the load-bearing open question: docs.neostack.dev states a floor of UE 5.4;

both betide.studio and neostack.dev marketing consistently say **"works on Unreal Engine 5.5, 5.6,

and 5.7"** (macOS and Windows). Every search targeted at UE 5.8 support turned up nothing — no

changelog entry, no forum post, no Discord reference, no version bump — confirming or denying it.

Given UE 5.8 shipped 2026-06-17 and the newest confirmed NeoStack version data point (1.0.74) is

dated 2026-04-24 (pre-5.8), **treat 5.8 compatibility as unconfirmed, not assumed-yes and not

assumed-no.** This is a direct pre-flight check item, not a research-answerable fact from open web

sources as of this date.

6. Licensing and pricing for a solo commercial developer

Two reported price points that at first looked contradictory reconcile cleanly once you separate the

core plugin from the cloud suite:

• Core plugin: one-time purchase, ~$110. "No monthly credits, no token middlemen, no expiring

balances... bring your own AI subscriptions." Covers the editor-agent-bridge functionality itself

(§3, §4 minus the Cloud-tier row). VERIFIED, betide.studio/neostack

and the Betide comparison page (which prices it explicitly

against competitors' $25-375/mo recurring models).

• NeoStack Cloud: $20/month, optional, adds hosted IDE, asset-aware version control with binary

support, CI/CD (automated UE5 builds/releases), AI code review, build/crash analytics, and a

10x-larger free-tier cloud-generation allowance. Studio workspaces need one paid seat per teammate

for Cloud features specifically — irrelevant for a solo dev.

• Fab marketplace terms: standard Fab Standard License, gated only by trailing-12-month gross

commercial revenue — Personal tier under $100k/yr, Professional tier above — both tiers grant

"the same scope of rights," just at different price points set by the publisher. VERIFIED,

Fab licensing docs. A

solo developer under the $100k threshold buys Personal tier with no restriction on shipping a

commercial game. No NeoStack-specific carve-out or extra commercial-use clause was found beyond

standard Fab terms.

• Direct purchase via neostack.dev is also available outside Fab (same plugin, per the download

page), which matters only if avoiding Epic's revenue share is a consideration — INFERRED, not

explicitly stated in what was fetched.

Net for this project: ~$110 one-time + whatever Claude Code subscription/API spend the project

already carries. The Cloud tier's CI/CD and version-control features overlap with things this repo

likely already does its own way (git, harness gates) — skippable.

7. Alternatives in the same class, mid-2026

Betide Studio's own comparison page names three direct competitors; independent research adds Epic's

own new native tooling and the open-source MCP-server ecosystem. Ranked roughly by

relevance-to-this-project (Claude-driven, UE 5.8-targeting, solo-commercial):

GitHub star/activity counts VERIFIED directly via the GitHub API (api.github.com/repos/...) on the

research date, not from search snippets.

8. Epic's own native UE 5.8 MCP plugin, in detail (the critical disambiguation)

This deserves its own section because it is the single easiest thing to conflate with NeoStack, and

this repo's own docs currently do (see §9).

UE 5.8's release notes ship a section literally titled "MCP Server (Experimental)": "With Unreal

Engine 5.8, we bring a new experimental MCP (Model Context Protocol) plugin for the Unreal Editor.

This plugin enables agentic AI systems to connect to the Unreal Editor" — VERIFIED,

Epic UE 5.8 release notes.

Full technical detail from the dedicated doc page

(Unreal MCP in Unreal Editor):

• Setup: Edit → Plugins → search MCP → enable → restart; toolsets come from a separate

"AllToolsets" plugin; generate agent client config via the

ModelContextProtocol.GenerateClientConfig console command.

• Transport: HTTP and Server-Sent Events only — no stdio, no WebSocket. Default endpoint

http://127.0.0.1:8000/mcp (configurable). No authentication layer — "the plugin is not safe

to expose beyond the local machine," explicitly localhost-only by design.

• Shipped toolsets: SceneTools, ActorTools, MaterialInstanceTools, ObjectTools (all

Python-authored), plus an experimental AttributeSetToolset for GAS, disabled by default. That is

the entire out-of-box coverage — **spawning actors, configuring lighting, creating material

instances, inspecting Slate widgets, running automation tests.**

• What it does NOT ship: no Blueprint-graph authoring toolset, no PCG toolset, no Landscape

toolset, no Sequencer toolset, no Niagara toolset — none of these are mentioned anywhere in Epic's

own documentation of the feature.

• Explicit caveats, quoted directly: "Learn to use this Experimental feature, but use caution

when shipping with it," and "many features are incomplete or missing. APIs and data formats are

subject to change at any time as it matures."

This is a genuinely different product from NeoStack — free vs. paid, first-party vs.

third-party, narrow-scene-editing vs. broad-asset-graph-authoring, and "experimental, expect breaking

changes" vs. a 1.0.x-versioned commercial product with a paying user base. It is real, it is exciting

context for UE's medium-term direction (Epic's own UE6 plans, announced the same June 2026 event,

describe first-class MCP support for Claude/Gemini/Codex as a permanent architecture going forward —

VERIFIED via multiple June 2026 industry write-ups), but **as shipped in 5.8 today it cannot replace

NeoStack for Blueprint/Niagara/Sequencer authoring** — it simply doesn't have those tools yet.

9. Cross-check against this repo's existing docs — where the naming already blurs

Reading the actual contract text (not just trusting version labels, per this repo's own

verification discipline) surfaced a real, evidence-based tension across three docs:

• T1_Build_Pipeline_Contracts §1.14 (the authoritative contract) names "NeoStack" specifically,

pinned to UE 5.7, and lists a rich tool-category set: project navigation, Blueprint ops, asset

ops, Niagara ops, Cinematic Sequencer ops, **World Partition ops, MetaHuman ops, Cesium ops, Gaea

import ops, SRTM heightmap import ops**, save-state ops, quest-system ops, combat-system ops,

build ops, and visual-feedback/screenshot-compare ops. Of these, §4 above independently confirmed

Blueprint, asset, Niagara, Sequencer, and screenshot/viewport-capture against NeoStack's own public

documentation. **World Partition, Cesium, Gaea, SRTM, and MetaHuman-Creator-invocation are not

claimed anywhere in NeoStack's own marketing or docs** — these may be reachable indirectly through

NeoStack's general asset/actor-manipulation tools, or may need a different pipeline leg entirely

(native Python/Remote Control, per this brief's verdict in §10), but they are not verified NeoStack

first-class tool categories as the contract's phrasing implies. The doc itself already hedges

correctly ("tool inventory... lock at NeoStack plugin documentation, not at this contract") — the

hedge is right; the specific category list should be treated as **the required capability list for

§1.4.1 to work**, not as a verified NeoStack feature inventory, until checked against

docs.neostack.dev directly.

• ROADMAP.md §6 (pinned 2026-07-03, i.e., current) correctly treats this as **Epic's own

engine feature**, grouped under "Engine" alongside Nanite Foliage/Mesh Terrain/MetaHuman

Crowds+Mass/Mover, all tagged Experimental, with well-calibrated hedging: **"MCP Server plugin —

Claude Code can eventually drive the editor"** (future tense, appropriately cautious — matches §8's

finding that it's real-but-narrow-and-experimental).

• MILESTONE_BUILD_PIPELINE.md (§ "Build-time authoring assist") instead attributes to **"the UE

5.8 MCP Server plugin"** a capability set — "Blueprint authoring, Niagara config, Sequencer

assembly, MetaHuman brief application" — that matches NeoStack's confirmed feature set (§4), not

Epic's actual shipped native plugin's feature set (§8: Scene/Actor/Material/Object tools only, no

Blueprint/Niagara/Sequencer/MetaHuman). Read charitably, this is describing NeoStack accessed via

its MCP-server connection mode (§3) running against the UE 5.8 target — a legitimate reading, since

NeoStack does offer an MCP mode. Read literally, the phrase "UE 5.8 MCP Server plugin" is Epic's own

product's exact name for its own exact feature, and a future reader (or an autonomous agent

parsing this doc without this context) could easily execute against the wrong tool and hit a

toolset gap it doesn't expect.

• PIPELINE_REVIEW_MISSION.md (2026-07-15, today, Josh's own stated ground truth) resolves the

ambiguity going forward: "Unreal work: Neostack AI. Engine target: UE 5.8" — explicitly flagged

as needing "the correction pass pins 5.8 and re-checks per-subsystem API surfaces." This brief is

that re-check.

Recommendation for the correction pass (not applied by this brief — that's Phase D's job, this is

research input): rename every "UE 5.8 MCP Server plugin" reference in MILESTONE_BUILD_PIPELINE.md

to disambiguate explicitly — e.g. "NeoStack (via its MCP-server connection mode, targeting the UE 5.8

build)" if that's what was meant — versus leaving it readable as Epic's same-named native plugin,

which cannot do the listed operations yet per §8. Trim T1_Build_Pipeline_Contracts §1.14's tool

list to what's actually confirmed (§4) plus a flagged-uncertain sublist (World

Partition/Cesium/Gaea/SRTM/MetaHuman) pending a direct docs.neostack.dev check once 5.8 compatibility

is confirmed.

10. Verdict — is NeoStack the right seam, and what goes where

Yes, keep NeoStack as the pinned Claude↔UE5-editor tool surface (Q1: confirm-same). It is real,

active, Claude-Code-native, commercially unrestricted for a solo dev, and functionally matches

exactly what the contract docs already describe. The two things that need to happen before relying

on it for the Phase 6 vertical slice are procedural, not a reason to reconsider the pick: (1) confirm

actual UE 5.8 compatibility directly (install it against a 5.8 project — this is a five-minute

pre-flight check this research cannot do from open web sources), and (2) resolve the §9 naming blur

so "NeoStack" and "Epic's native MCP" never get silently swapped for each other in a doc an

autonomous agent will execute against later.

**What belongs on NeoStack (or an agent-in-the-editor bridge generally) vs. UE's native Python /

Remote Control APIs** — this is the real architectural question underneath Q1, and it maps cleanly

onto this project's own model-tiering doctrine (deterministic-first, premium-model spend reserved for

judgment calls):

• NeoStack's seam = interpretation, not transcription. Anywhere turning canon *intent* into an

engine *asset graph* requires judgment calls with no 1:1 mapping already specified — first-draft

Blueprint logic from a quest description, a Behavior Tree from a boss's described verb-set, a

Sequencer cinematic from a scene's prose beats, wiring a Niagara effect to match a described vril

signature. This is exactly the register NeoStack is built for (conversational, live-edited,

undo-backed) and exactly where Claude Code is independently confirmed to perform best driving it

(§5).

• Native Python / Remote Control's seam = bulk, structured, already-specified. Anywhere the

source-to-engine mapping is already fully known and repeats across many rows — importing

T0_Weapon_Registry/T0_Creature_Roster rows into Data Tables at volume, batch-placing actors from

a manifest, running the same operation across all 69 regions, CI-style repeatable builds and

automated tests. This is more reliable (no per-call LLM non-determinism), cheaper (no token spend

per row), versionable, and rerunnable — and it's the same split this project's own global working

agreement already states in the abstract ("deterministic gates + scripts first, the cheapest model

that works, premium model for judgment/adjudication only"). Epic's own native UE 5.8 MCP plugin

(§8) is arguably a *third*, narrower option for exactly this bulk/scene-level tier once it matures

past Experimental — its Scene/Actor/Material/Object toolset is a plausible free substitute for some

of what a script would otherwise do via raw Python, but it is not a Blueprint/Niagara/Sequencer

substitute for NeoStack today.

• **Boundary items flagged uncertain in §9 (World Partition, Cesium, Gaea, SRTM, MetaHuman Creator

invocation) are, on current evidence, better assumed to be native pipeline legs** (Cesium for

Unreal and Gaea2Unreal are themselves already named as separate pipeline contracts in this repo,

each with their own plugin/import path) rather than NeoStack chat-tool operations, **until a direct

docs.neostack.dev check says otherwise.**

What the Translation Doc needs to specify, concretely:

1. Confirmed UE version pin — do not write the Translation Doc against "UE 5.8 + NeoStack"

until a direct install test confirms compatibility; if it lags, the fallback is either staying on

UE 5.7 for the NeoStack-mediated legs, or substituting the self-hosted

chongdashu/unreal-mcp-class open-source bridge (most mature/active OSS option found, §7) for

whatever NeoStack can't yet reach on 5.8.

2. Agent/transport pin — Claude Code as the driving agent (independently corroborated as the

best-performing choice, §5), and an explicit choice between ACP mode and MCP-server mode for how

Claude Code reaches NeoStack (both are supported; the contract doc doesn't currently pick one).

3. A hard operation-routing boundary list — which asset types/operations go through NeoStack's

conversational tool-calls vs. which go through deterministic Python/Remote-Control scripts, so an

autonomous build loop never asks an LLM agent to hand-author something a script would do more

reliably at volume (e.g., importing all 72 weapon rows one conversational turn at a time).

4. A human-in-the-loop checkpoint policy — NeoStack's only built-in safety net is undo, not

pre-approval; the Translation Doc should define what triggers a mandatory pause (e.g., any

operation touching hard-line/§17 content, any bulk/destructive operation) since the tool itself

won't enforce that boundary.

5. The licensing/cost line — ~$110 one-time (Personal tier via Fab or direct purchase) +

whatever Claude Code subscription the project already runs; the $20/mo Cloud tier is optional and

likely skippable given this repo already has its own git/harness-based CI equivalent.

6. A named fallback — if UE 5.8 support doesn't land in time, chongdashu/unreal-mcp (2,034

GitHub stars, actively maintained, free, self-hosted MCP server explicitly aimed at Claude-class

clients) is the best-corroborated open-source substitute found in this pass, and Epic's own native

MCP plugin (§8) is the free first-party fallback for the narrower Scene/Actor/Material tier

specifically.

---

Sources consulted (primary and corroborating)

• neostack.dev — official product site
• docs.neostack.dev — official docs
• betide.studio/neostack — maker's product page (fetched directly,

quoted verbatim for ACP/MCP and pricing details)

• aik.betide.studio/comparison — Betide's own

competitor-comparison page

• neostack.dev/download
• fab.com listing 0e725daa-... —

direct fetch blocked (HTTP 403); corroborated via search-result snippets instead (version/date data

points used with that caveat)

• Philip Conrod — NeoStack AI hands-on, 2026-02-09
• neodocs.betide.studio/docs — NeoAI (sibling/distinct product)
• Betide Studio on X — NeoAI announcement
• Tracxn — Betide Studio company profile
• Epic Developer Community — UE 5.8 release notes
• Epic Developer Community — Unreal MCP in Unreal Editor
• Epic Developer Community — Fab licenses and pricing
• tryaura.dev and Epic forums — Aura showcase thread
• ludusengine.com and docs.ludusengine.com/plans-and-billing
• github.com/chongdashu/unreal-mcp,

github.com/ChiR24/Unreal_mcp,

github.com/remiphilippe/mcp-unreal,

github.com/runreal/unreal-mcp — star counts/activity dates

pulled directly via the GitHub API on the research date

• In-repo cross-references: _source/01_Tier_1_Foundation/T1_Build_Pipeline_Contracts [ACTIVE v1.0].md

§1.14/§5, docs/ROADMAP.md §6, docs/MILESTONE_BUILD_PIPELINE.md (build-time authoring assist

section), docs/PIPELINE_REVIEW_MISSION.md (stack facts, 2026-07-15), docs/pipeline_review/GAP_MAP.md

(L20/L175, L200, L368-369), docs/pipeline_review/STACK_FACTS_QUESTIONS.md Q1

Generated by harness/site/structure_site.py — the URL path is the repo path. review root