# UE MCP Strategy

## Purpose

Learn the landscape around a single strategic question — **should Unreal Engine open an MCP (an agent-facing tool surface) rather than build its own in-house AI "harness," and does doing so strengthen the runtime's gravity?** — well enough to reason about it critically rather than just intuitively. This workspace exists because the operator instinct ("Epic captures value at the runtime, not the tooling, so opening the tooling is free upside") is probably right but is currently un-grounded in how the runtime, MCP, monetization, and the AI-content-generation frontier actually work.

## In Scope

- How the Unreal Engine runtime + editor actually work, at the level that determines whether an MCP is feasible (reflection/UObject system, Blueprints, editor-vs-runtime split, the cook/build pipeline, Verse/UEFN).
- What MCP (Model Context Protocol) is, how it exposes tooling to models, and the state of existing Unreal MCP servers and UE's programmatic surfaces (Python editor scripting, Remote Control API).
- The build-vs-expose decision: an opinionated in-house **harness** vs. an open **MCP**, and how comparable platform companies (Figma, Blender, Unity, Adobe, Notion, GitHub) have chosen.
- Where Epic actually captures value (UE royalty, Fortnite, UEFN creator economy, Fab) and how AI-generated content does or does not flow through those channels.
- The procedural + AI content-generation frontier (UE5 PCG framework, text-to-3D, world/scene generation) and how good "models at PCG" really are today.
- What "runtime gravity" / moat / lock-in mechanically means, and the strongest counter-thesis.

## Out of Scope

- Epic's internal roadmap or confidential plans (reason from public facts only).
- Deep rendering/physics internals beyond what bears on the decision.
- A general survey of all MCP servers or all AI tooling.
- Building anything — this is a learn-and-decide workspace, not an implementation.

## Current Thesis / Working Questions

- **Working thesis:** Open an MCP, not a harness — Epic captures value at the runtime, so the authoring layer is free to commoditize; an MCP lets Epic rent the frontier labs' intelligence while every generated asset deepens UE's content gravity. The one rule: the MCP must emit *UE-advantaged* content, never engine-neutral content.
- Is the value-capture claim actually true — does the 5% royalty capture the AI-content wave, or does that value flow through Fortnite/UEFN engagement instead? (If the latter, "runtime gravity" is really "Fortnite-ecosystem gravity," and a different team owns the decision.)
- Is an MCP technically cheap because UE is already introspectable (reflection system, Python/Remote Control API), or is that an over-simplification?
- What is the real failure mode — does an engine-agnostic MCP standard commoditize UE into a swappable backend?
- How good is AI procedural generation *today*, and does exposing tooling actually improve models at it, or is that wishful?

## Output Goals

- Durable concept notes that explain each layer (runtime, MCP, value capture, AI gen, gravity) in plain terms.
- Cited source digests so every load-bearing claim is auditable.
- A first-pass synthesis that states the decision frame and a critical-open-questions note that lays out where the thesis breaks and what to resolve next.
- Seeds for a possible `ideas/` note / post — leading from the operator edge (where value lands), not consensus AI-hype takes.

## Promotion Candidates

- Value-capture layer determines who can safely open an MCP
- Rent-the-frontier's-intelligence vs own-the-intelligence
- Content gravity (not the runtime) is the real lock-in
- UE-advantaged vs engine-neutral generation

## Known Tensions

- Opening the authoring layer is free upside *only if* value capture really sits at the runtime; if it sits at engagement, the calculus shifts.
- An MCP lets you free-ride frontier models, but it also invites an engine-agnostic standard that could commoditize the runtime.
- "Models get better at PCG as we expose tooling" is an attractive flywheel claim that may be more hope than mechanism — needs evidence.
- A thin harness may still be needed for the median creator even if the MCP is the strategic bet; either/or framing may be false.