Skip to content

Quick Start

The common workflow: ask Claude to create a Blueprint, then verify the result.

  • Blueprint AI installed and the editor launched at least once (so the Claude skill is registered)
  • Claude installed locally — or any other AI tool that can shell out to the Blueprint AI CLI

Open Claude — or whichever assistant you set up in AI Service Setup — and ask it to create a Blueprint. For example:

Create a Blueprint Actor called BP_Pickup in /Game/Pickups. Add a StaticMeshComponent as the root and a float variable called Value with a default of 10.

The AI will:

  1. Look up the schema for create_blueprint, add_component, and add_variable
  2. Generate a JSON operations plan
  3. Call the Blueprint AI CLI to validate and execute it

The result appears in your Content Browser at /Game/Pickups/BP_Pickup.

You can also call the CLI yourself to explore or test:

Terminal window
CLI="python Plugins/BlueprintAI/Scripts/cli.py"
# List all available operations
$CLI schema
# See the full schema for one operation
$CLI schema create_blueprint
# Inspect an existing asset
$CLI inspect /Game/Pickups/BP_Pickup
# Search the Blueprint action database
$CLI search "PrintString"

Operations are JSON arrays sent via stdin:

Terminal window
echo '[{"op": "create_blueprint", "path": "/Game/Pickups", "name": "BP_Pickup", "parent": "Actor"}]' \
| python Plugins/BlueprintAI/Scripts/cli.py apply --stdin

When you run any operation:

  1. The CLI validates the JSON against the operation schema
  2. It resolves dependencies (e.g. ensures the parent class exists)
  3. It generates Python code and sends it to the engine via a UDP remote execution channel
  4. The engine executes it, compiles, and saves the asset
  5. The CLI reports success or a structured error back to the AI

If the editor is not open, Blueprint AI launches it in commandlet mode automatically.