Skip to content

FAQ

No. Direktor is not an AI itself, and it doesn’t include or provide free access to any AI or LLM model. It’s an Unreal Engine plugin that bridges the editor to the AI client of your choice — Claude Desktop, Claude Code, Cursor, and many more.

You bring your own AI client. Free tiers of some clients work for many tasks; heavier automated use follows your AI provider’s own pricing. Direktor itself is a one-time purchase.

run_python is disabled by default. When you opt in, the AI executes Python in your editor’s process, so anything you can do in Python it can do too — keep your project under version control. Console commands are enabled by default, but Unreal blocks the destructive ones in the editor context. Most other tool calls are recorded as a single undo step you can revert.

Direktor runs entirely in-process and never makes outbound calls. Your AI client may need internet access, but that’s between you and your AI provider.

No. Direktor is idle when no client is connected. Tool calls run on the game thread one at a time, and the reflection catalog builds once at startup (cached after the first run) with a small memory footprint.

No. Direktor is editor-only. It does not run in a shipped or packaged game.

Unreal Engine 5.0 through 5.8. Some advanced tools are version-gated (Niagara and Behavior Tree authoring need 5.3+, geometry and PCG need 5.4+) and return a guided message on older versions.

No. The copy you install from Fab is precompiled per engine version, so Direktor works in any project, including Blueprint-only ones.

Any MCP-compatible client. The bridge (unreal-direktor-mcp) works everywhere; clients that speak Streamable HTTP directly can connect to http://127.0.0.1:13579/mcp without it. See Quick Setup for per-client configuration.