Quickstart

Once the plugin is enabled and your agent is connected, you don't run RPCs by hand — you ask your agent in plain language, and it discovers and calls them. Here's the loop, end to end.

1. Confirm the connection

With the Unreal Editor open and the plugin enabled, your agent should list a tool named call from the editor-automation server. If it doesn't, see Troubleshooting.

2. Let the agent discover the surface

The agent finds what it needs through the same tool. An empty call() returns the namespace index; naming a namespace returns that page:

call()
→ actor · asset · blueprint · material · niagara · widget · … (50+ namespaces)

call("blueprint")
→ blueprint.create · blueprint.compile_bpir · blueprint.decompile_bpir · …

3. Ask for something real

Prompt your agent naturally — for example:

• “Create a Blueprint actor that prints a message on BeginPlay, waits a second, then prints again.”
• “Make a material with a flat blue base color and save it under /Game/Materials.”
• “Show me what this Niagara system does so I can review the change.”

The agent translates that into a call. For the first prompt it compiles a Blueprint graph from text (BPIR):

call("blueprint.compile_bpir", { text: "entry event BeginPlay() {
    call PrintString(InString: \"Hello World\")
    call Delay(Duration: 1.0)
    call PrintString(InString: \"Done\")
} " })
✓ compiled · 1 event graph · saved

4. Verify and save

A good agent reads back what it changed before saving — dumping the asset, decompiling the graph to text, or capturing a screenshot — then saves explicitly. Because the graph round-trips to text, you can review the diff in git like any other change.

Next: Concepts →