Docs · Extensions · Unreal Engine

AI PCG

Procedural Content Generation authoring — build PCG graphs (nodes, connections, settings, graph parameters), place them on PCG volumes, generate, and read back instance counts.

Last updated: 2026-11-15 View as Markdown

At a glance

Capabilities
4 capabilities
Version
0.3.0
Package id
com.ivanmurzak.unreal-ai-pcg
Engine
Unreal Engine · Unreal Engine 5.7+
Wraps
PCG
Repository

What the agent can do with it

Adds no MCP tools: it contributes an RQL module with the pcgGraphs, pcgNodeTypes, pcgTemplates, pcg sources and adapters for PCGGraph, PCGNode, PCGComponent. That powers 4 capabilities the agent reaches through rql-query / rql-mutate. How RQL works →

  • Author graphs

    Create PCG graphs (optionally from a template), add and remove nodes from the palette, and connect or disconnect their pins.

    pcg | call AddNode("/Game/PCG/PCG_Forest", "Surface Sampler")
  • Settings & parameters

    Set node settings as text (meshes with weights, filters, attributes), and add graph parameters bound to node settings.

    pcg | call SetParameter("/Game/PCG/PCG_Forest", "Density", "0.02")
  • Place & generate

    Place a graph on a PCG volume or an actor’s PCG component, generate with a seed and read back instance counts, or clean the generated content up.

    pcg | call Generate("Forest", 7)
  • Inspect graphs

    List PCG graphs, the node types the palette offers (with descriptions and pins) and templates, and read a graph’s nodes, edges and parameters.

    pcgNodeTypes | where Category == "Filter" | select Name Type Description Inputs Outputs

See it in the Unreal Engine catalog →

Install

Requires the base plugin UnrealMCP 0.26.0 or later. The extension ships as C++ source: rebuild the project (or pass --build) before opening the editor; then the RQL module loads with the package — the agent sees it in rql-schema.

unreal-mcp-cliunreal-mcp-cli install-extension com.ivanmurzak.unreal-ai-pcg <UEProject>

When to ask the AI for it

Ask for procedural scattering — a forest, rocks or props over a landscape, filtered by slope and height — without writing your own C++ or Python generator.

Not sure if it's installed? Ask the agent to run rql-schema — the pcgGraphs source only exists when the package is present.

Example prompts

Copy-pasteable first requests. Start with one, verify it in the editor, then build up.

  1. Create a PCG graph PCG_Forest that samples the landscape surface and spawns SM_Pine and SM_Birch at 3:1, with a Density parameter.
  2. Place PCG_Forest on a 500 m volume called Forest and generate it with seed 7; how many trees did it spawn?
  3. Which PCG filter nodes are there?

Keep going