cu Commands

cu exposes nine subcommands. Three are local-only (read the bundled manifest or local files) and six hit the TCP bridge.

Subcommand	TCP	Purpose
cu list	—	List manifest commands (optionally filtered)
cu help	—	Full reference for one command
cu search	—	Fuzzy search the manifest
cu log	—	Read filtered editor / PIE log (local file)
cu ping	✅	Probe the bridge, report latency
cu doctor	✅	Diagnose CLI / manifest / editor state
cu exec	✅	Send a command to the bridge
cu preview	✅	Open an asset editor and screenshot its viewport
cu catalog	✅	Build a static HTML asset catalog of /Game

cu list

cu list [--category <id>] [--limit N]

Reads manifest.json; emits one line per command:

spawn_actor                              [scene_actors]  Create a new actor in the current level.
delete_actor                             [scene_actors]  Remove an actor from the level by name.
...

--limit 0 (or omitted) lists all matches.

cu help <command>

cu help spawn_actor

Output sections: full description, params (typed --flag form with required vs optional), and a runnable cu exec example.

This is the same content unreal_doc(<command>) returns over MCP — the two paths are interchangeable.

cu search <query>

cu search "focal length"
cu search foliage --category landscape_foliage --limit 5

In-memory keyword scoring (no embeddings):

Match	Score
Query is substring of command name	+10
Query token appears in command name tokens	+5 each
Query token appears in summary	+3 each
Query token == category id	+2 each

Sorted descending, capped at --limit (default 10).

cu ping

cu ping

Returns ✓ pong on success. Use --json for machine-parseable output.

cu doctor

cu doctor

Emits:

• Manifest path and command count
• bridge_commit recorded at manifest build time
• Bridge port and editor reachability (probes ping)

Useful first step when something stops working — pinpoints whether the manifest is stale, the editor is down, or the port is blocked.

cu exec <command>

Three input styles, in order of typical use:

# Typed flags — derived from manifest schema
cu exec spawn_actor --name MyCube --type StaticMeshActor --location 0,0,100

# JSON string — easier for nested params
cu exec spawn_actor --params-json '{"name":"MyCube","type":"StaticMeshActor","location":[0,0,100]}'

# JSON file — large or reused payloads
cu exec spawn_actor --params-file /tmp/spawn.json

If --params-json or --params-file is given, typed flags are ignored.

Note

Lists in typed flags use comma form (--location 0,0,100) for List[float]. Boolean flags are --simulate (true) / --no-simulate (false). Anything more complex — pass JSON directly.

cu log

cu log [--category <name>] [--verbosity <level>] [--max N] [--json]

Reads <ProjectDir>/Saved/Logs/<Project>.log locally — no TCP, works whether the editor is running or not. Filters by --category (e.g. LogTemp, LogBlueprintUserMessages) and --verbosity (all / info / warning / error). Returns the most recent --max matches (default 100).

Replaces the older get_runtime_log MCP tool — same semantics, no MCP token cost.

cu preview <asset_path>

cu preview /Game/Meshes/SM_MyMesh.SM_MyMesh [--out <png>]

Opens the asset editor and captures its viewport at editor quality — full PBR, Nanite, streamed textures, studio lighting. Single asset only, ~1–3s per call. Default output is a temp file keyed by the asset path; the PNG path is printed on stdout.

Currently static mesh only; other asset types follow.

cu catalog

cu catalog [--out DIR] [--path /Game/...] [--classes CSV]
           [--batch N] [--size N]

Walks the asset registry under --path (default /Game), extracts a thumbnail per asset, and emits a self-contained static HTML grid plus a machine-readable manifest:

<ProjectDir>/Saved/ClaudeUnreal/Catalog/
  index.html       ← grid + class chip filter + search + copy-path
  manifest.json    ← {generated_at, counts, assets:[{path,class,tier,png}]}
  thumbs/*.png

Two-tier extraction:

1. tier: cached — reads the FObjectThumbnail baked into the .uasset on save (the same image the Content Browser shows). Fast, covers ~90% of meshes / materials / textures.
2. tier: rendered — falls through to ThumbnailTools::RenderThumbnail for assets without a cached thumbnail (e.g. fresh Blueprints).

Assets without either path are recorded tier: icon-fallback (no PNG, grey card); per-asset failures become tier: error rows but never abort the run.

Default class whitelist: StaticMesh, SkeletalMesh, Material, MaterialInstanceConstant, Texture2D, ParticleSystem, NiagaraSystem, Blueprint (Blueprint additionally filtered to AActor-derived).

--batch controls how many asset paths each TCP call carries (default 50); per-batch progress lines go to stderr, the final file:// URL to stdout.

Hard fail if PIE is active — RenderThumbnail shares the GPU and won’t run concurrently.

cu catalog --path /Game/Megascans --batch 50
# stderr: [catalog] 200/847 (cached=187 rendered=11 fallback=2 error=0)
# stdout: file:///.../Saved/ClaudeUnreal/Catalog/index.html

Global flags

These work on any subcommand:

Flag	Purpose
--port N	Bridge port (default 55557)
--json	Emit JSON to stdout instead of human-readable lines
--quiet	Suppress stdout; report via exit code only

Next

• Discovery flow — find the right command when you don’t know its name.
• Reference — the full per-category command index.