Triardor Devkit gives AI developers direct, verified control of a running Unreal Engine 5.8 editor. They create Blueprints, materials, AI, abilities, animation, UI and levels, then check that every change really happened.
Every custom tool, editor bridge, knowledge base and AI setting lives in one place, so every machine and every AI assistant uses the same version of every tool, and nothing drifts.
The Unreal MCP drives an open UE 5.8 editor: authoring, compiling, playing (PIE) and taking screenshots.
4,282 Unreal tutorial videos turned into searchable recipes, plus 22 evidence-backed gameplay guides that agents follow step by step.
Every edit is guarded, logged, verified and can be undone. The AI proves its work instead of claiming it.
Tools are defined once and work the same way for every AI assistant on every machine.
A generated map answers "I want to X → use Y" across 400 components, each marked with how well it has been proven.
A new workstation gets the complete toolkit in one guided setup.
The Model Context Protocol (MCP) is the standard way AI agents call tools. Our server runs each tool call inside the Unreal editor, then reads the result back.
Grouped into tool families. Each family can author, inspect and verify.
create · components · variables · interfaces · compile
events · calls · casts · flow control · pins · node text import/export
expressions · connections · instances · parameter collections · compile errors
ability system · abilities · effects · modifiers · tags
behaviour trees · blackboards · EQS · navmesh · StateTree
anim blueprints · state machines · transitions · Control Rig · skeletons
widget blueprints · buttons · text · event binding · viewport
spawn · move · tag · folders · level management
Enhanced Input actions/mappings · DataTables (JSON/CSV)
level sequences · tracks · keyframes · playback ranges
PIE control · console · screenshots · diagnostics · batch
API discovery + a bridge to Epic's own ~248 engine tools
An illustrative session: the agent looks up how to build the mechanic, builds it in the live editor, compiles it, plays it and checks the result. Each step reports what really changed.
Most AI-to-editor bridges report "done" whether or not the change happened. Every action through ours passes five guards.
Edits are only allowed when editing is explicitly switched on.
The editor side is checked for compatibility before any command runs.
Every action is recorded, so nothing happens without a trail.
After each edit, the server reads the editor again to confirm the change landed. Compile errors are shown, never hidden.
Edits can be reversed, and it reports exactly what was changed.
Live regression suite passing against a real running editor.
Known quality gaps went to zero, and a ratchet keeps them from coming back.
Turns Blueprints into readable summaries and meaningful diffs, for review and explanation.
Where we started from open-source work, we rebuilt rather than copied. Our version compiles faster, shows errors instead of hiding them, explains failed connections precisely, reads back every write, and fixes bugs present in the original.
We turned thousands of Unreal Engine tutorial videos into a searchable knowledge base. The AI looks up proven recipes, then follows full build pipelines from zero to a finished feature.
Each guide gives the build order and the common pitfalls, with sources:
Plus project harvesting: lessons from finished projects become reusable knowledge for the next one.
The infrastructure work is done and tested. What limits output now is AI usage. Every extra token goes into editor actions that are verified and reviewed, not into chat.