# Gameplay & AI

An enemy that fights back.

Hand your AI a JSON description of the behavior you want, and it just builds the whole thing
— behavior trees, ability systems, state machines, combo chains, the input that drives
them — then plays the game to prove it actually works.

## Game AI (230 actions)

*The most comprehensive AI tooling on any MCP server.*

Full lifecycle for Behavior Trees, Blackboards, State Trees, EQS, Smart Objects, AI
Controllers, Perception and Navigation — including build_behavior_tree_from_spec and
build_state_tree_from_spec, which turn a JSON behavior description into a finished asset.
Scaffolds patrol, combat and companion AI, then debugs it live during PIE instead of just
hoping it works.

Covers: Behavior trees, State trees, EQS, Smart objects, Perception, Build-from-spec.

## Gameplay Ability System (135 actions)

*Abilities, attributes, effects — wired.*

Creates abilities with activation policies, cooldowns, costs and tags; full AttributeSet
CRUD; Gameplay Effect authoring with modifiers, stacking and periods; ASC management, cues,
targeting and input binding — plus scaffolds that spin up a complete GAS setup from a
template instead of you wiring it by hand.

Covers: Abilities, Attribute sets, Gameplay effects, Cues & tags, Targeting, Scaffolds.

## Logic Driver State Machines (66 actions)

*Game flow as a spec.*

Full Logic Driver Pro integration: state machine CRUD, graph read/write, transition rules,
runtime PIE control and one-shot build_sm_from_spec — with ready-made scaffolds for doors,
health systems, patrols, dialogue trees, elevators, puzzles and inventories, so you're not
starting from a blank graph.

Covers: FSM CRUD, Transition rules, Runtime control, Dialogue & quest scaffolds.

## Combo Graphs (13 actions)

*Attack chains, choreographed.*

Creates and edits ComboGraph combo trees — nodes with montage animations, edges, effects
and cues — and scaffolds combo abilities that bridge ComboGraph straight into the Gameplay
Ability System.

Covers: Combo trees, Montage nodes, GAS bridge.

## Enhanced Input (4 actions)

*WASD, wired in four calls.*

Creates Input Actions and Input Mapping Contexts as real assets, then binds keys to them
with the full modifier and trigger stack — negate and swizzle for a proper movement stick,
hold, tap and pulse for everything else, keyboard, mouse and gamepad alike. Reads a context
back so your AI can verify what it just mapped instead of assuming. The matching event nodes
and the subsystem registration get authored into your character Blueprint alongside it.

Covers: Input Actions, Mapping Contexts, Key mapping, Modifiers & triggers, Gamepad & mouse,
Read-back verification.

## PIE Input Simulation (14 actions)

*Your AI plays the game to prove it works.*

Drives a running Play-In-Editor session the way a player would — press keys, click, drag,
type, scroll. The last step of autonomous verification: build the feature, then actually go
play it. It can also fire an Input Action straight through Enhanced Input, running your real
modifiers and triggers, which means an ability can be tested before anyone has bound a key
to it. Camera aim is set with an optional hold so it survives a per-tick camera system that
would otherwise overwrite it on the next frame, and your AI can detach to a free-fly
spectator to go look at something, then hand control back.

Covers: Key & mouse simulation, Enhanced Input injection, Control rotation hold, Spectator
free-fly, Autonomous verification.

---

One of 7 skill categories in Shovel, an Unreal Engine editor plugin and MCP server. HTML
version: https://shovel.to/gameplay-ai/. Site index for agents: https://shovel.to/llms.txt.