Tools Reference — Overview
ClaudusBridge exposes 567+ tools organized into 50+ categories. Each tool is a callable MCP function that executes an action in the Unreal Editor.
The live count is read at runtime from Router::GetTotalToolCount() and threaded through every generated artifact (this docs page, the in-editor dashboard, the auto-generated Saved/ClaudusBridge/CLAUDE.md, the /health endpoint, the JSON-RPC serverInfo). When the plugin is updated, every count stays in sync automatically — there are no hardcoded magic numbers to drift.
Categories at a Glance
| Category | Tools | Description |
|---|---|---|
| Blueprint Nodes | 18 | Add, delete, configure, batch edit, auto-layout graph nodes |
| Pin Connections | 3 | Connect, disconnect, disconnect all pins |
| Events | 2 | Standard events (BeginPlay, Tick) and custom events |
| Variables | 3 | Create, delete, set properties on Blueprint variables |
| Functions | 5 | Create, delete, rename functions; add inputs/outputs |
| Inspection | 7 | Read Blueprints, analyze graphs, compile, describe, reparent |
| Actors | 23 | Spawn, delete, transform, properties, components, line trace, attach |
| Assets | 17 | Find, duplicate, rename, delete, import, export assets |
| Materials | 30 | Create materials, expressions, instances; connect, recompile |
| Editor | 48 | Viewport, PIE, capture_editor_window, console, navigation, vision, cognitive map, scan_360, proximity_alert |
| World | 9 | Levels, lights, physics, gravity |
| Validation | 5 | Validate Blueprints, search across BPs, find unused vars |
| Behavior Trees | 9 | Create, inspect, modify AI behavior trees and blackboards |
| Input | 6 | Enhanced Input actions, mapping contexts, key bindings |
| Utilities | 10 | Output log, Python exec, undo/redo, transactions, shader paths |
| Snapshots | 4 | Create, restore, list, delete backup snapshots |
| Data Types | 6 | Create structs, enums; add fields/entries |
| Animation | 10 | Sequences, notifies, curves, bones, montages, blend spaces |
| AnimGraph | 1 | Add AnimGraph nodes (BlendSpacePlayer, SequencePlayer) |
| Mesh | 12 | Static/skeletal mesh info, Nanite, collision, LODs, materials |
| Audio | 8 | List sounds, play/stop, spawn audio components |
| Widgets | 29 | Create UMG widgets, palette, layout, anchors, style, reflector, events |
| Sequencer | 12 | Level sequences, bindings, tracks, keyframes, playback |
| Gameplay | 8 | Gameplay tags, navigation mesh, pathfinding |
| Physics | 8 | Physics assets, collision, mass, forces, impulses |
| Level | 10 | Sublevels, streaming, build all, save, dirty packages |
| Source Control | 8 | Status, checkout, checkin, revert, history |
| Landscape | 11 | Terrain info, layers, foliage, heightmap, resolution |
| Media | 6 | Media players, playback control, state |
| Processes | 7 | Launch servers/clients, multiplayer PIE, process management |
| Build | 3 | full_build, clean_plugin, get_build_status |
| Runtime Inspector | 5 | Inspect actors and components during PIE gameplay |
| Performance | 3 | FPS snapshots, continuous monitoring |
| Test Runner | 4 | Automated PIE testing with condition checks |
| Niagara | 30 | Particle systems, emitters, modules, renderers, parameters |
| MetaSound | 15 | Procedural audio nodes, connections, preview |
| DataTable | 4 | Schema, rows, add/delete entries |
| MVVM | 8 | ViewModel bindings for UMG widgets |
| Procedural Mesh | 38 | Create geometry, booleans, extrude, sculpt, fracture, displace |
| Widget Tree | 5 | Wrap, unwrap, reparent, slot properties, compile |
| Asset Import | 9 | Import textures, FBX; export; set properties |
| BlueprintLisp | 4 | Text-based Blueprint authoring language |
| Editor Events | 4 | poll_editor_events, subscribe_events, publish_agent_event, record_feedback |
| Memory | 5 | memory_remember, memory_recall, memory_topics, memory_search, memory_consolidate |
| Components | 4 | Static mesh, physics, transform properties on BP components |
| Widget | 2 | add_child_widget, remove_widget_from_tree |
| Neural Cortex | 12 | ai_perceive, ai_navigate, ai_create, ai_modify, ai_inspect, ai_build, ai_scene, ai_blueprint, ai_media, ai_engineer, ai_evaluate, ai_brain_map |
| AI Vision | 7 | ai_vision, ai_vision_stream, ai_vision_identify, NNE GPU inference |
| Cognition (0.6.x) | 60+ | Workflow templates, multi-agent workflows, task plans, skills, agents, action queue, smart_start, workflow_timeline, workflow_checkpoints, continuation_stats, last_auth_state, observations, project intelligence, native tool/skill authoring, … |
What's New in 0.6.x
Cognition Tier (0.6.0)
ClaudusBridge now ships a dedicated Cognition category that consolidates the project-intelligence, observation, workflow, and self-programming tools. Highlights from 0.6.0:
claudus_checkpoint_workflow— Snapshot asset paths + last-modified timestamps for a workflow id underSaved/ClaudusBridge/WorkflowCheckpoints/<checkpoint_id>.json. Resolves the asset list in this order: explicitasset_paths→scan_pathsviafind_assets→ the workflow's persistedassets[]→ empty. Fires anauto_observationmemory entry taggedworkflow_checkpoint+<workflow_id>+<checkpoint_id>.claudus_list_workflow_checkpoints(workflow_id?, limit?)— Inventory of recorded checkpoints. Newest first. v1 ships the audit/diff layer; rollback is intentionally out of scope until source-control integration lands.claudus_continuation_stats— Projectssaved_inference_msfrom auto-continuation batching (BatchedCount × ~1500 ms by default) plus a coarse USD estimate. One tool call shows finance the savings.claudus_get_last_auth_state— Reads the sanitizedlast-auth-state.jsonwritten by the Node provider runtime (presence + last-modified timestamp per provider credential file; never token values) and returns a coarseverdict:ok/anthropic_missing_but_alt_present/needs_login.
Action-Failure Observations (0.6.0)
Every tool call that returns status != "success" now writes a memory entry tagged action_failed + tool:<name> + severity:<high|low> (and kind:unknown_command for missing tools). High severity is reserved for mutation-prefixed tools (create_/write_/delete_/save_/modify_/rebuild_/set_/ai_create_/ai_modify_/ai_build_); read-only failures log at low severity so the journal doesn't drown in trivial misses. Combined with claudus_recall_observations(tag=action_failed, since_hours=…) a fresh session gets an immediate audit of recent breakage.
Self-Healing Provider Pipeline (0.6.1 → 0.6.3)
Provider-runtime restarts no longer leave the editor stuck against a dead port:
- 0.6.1 added rediscovery to the dashboard auto-connect path (
ConnectProviderRuntimeAsync). NewFCBClaudusAI::TryRediscoverProviderRuntimeURL()readsdispatch.jsonand returns a freshly-built URL when the port differs. NewCBDesktopAuthBridge::InvalidateDispatch()drops every memoization keyed on the old port. Plus: arglesstools/callno longer spamsLogJsonwarnings. - 0.6.2 extended the same self-heal to both normal chat-send paths:
ResolveAndAppendReply(synchronous, used byask_claudus/submit_chat_sync) andResolveAndAppendReplyAsync(async, used bySubmitHumanMessagefrom the Output Log). - 0.6.3 flipped the policy from retry-after-failure to proactive refresh: a cheap
stat()ofdispatch.jsonat the entry of every chat request detects a runtime restart viamtimeand swapsBaseURLbefore the request goes out. NoLogHttpwarnings, no 2 s libcurl timeout against the dead port, no visible recovery message — just the right port from the start.
The retry-after-failure logic from 0.6.1/0.6.2 stays as a safety net.
What's New in 0.5.0
Editor Events Category (4 tools)
The single most impactful additions for AI-agent ergonomics:
poll_editor_events— Drain YOUR session's event queue. Replaces busy-polling state withlist_actors/describe_view/ etc.subscribe_events— Tell the editor which event types YOUR session wants. Subscriptions act as a filter at insert time, so you only ever see events you asked for. Pass["all"]to see everything.publish_agent_event— Agent-to-agent messaging primitive. Channel-based broadcast or targeted delivery on the same per-session queue.record_feedback— Persist a plugin-improvement signal toSaved/ClaudusBridge/feedback.jsonland broadcast it on theagent_messagebus on channelfeedback.<kind>. The canonical source of plugin self-improvement signal.
Together, these enable multi-agent swarm patterns (watcher, specialist, pair-programming, recorder, sentinel) and close the verification gap between "tool returned success" and "the visual reality matches the intent". See Multi-Agent Coordination for patterns.
Memory Category (5 tools, new)
The persistent shared knowledge base that lets every agent (Claude Code, Codex CLI, Cursor, Windsurf, …) accumulate and read tribal knowledge across sessions:
memory_remember— Append a learning toSaved/ClaudusBridge/memory/<topic>.mdand_log.jsonl. Topic names auto-normalize (lowercase, whitespace → underscores). Most-recent entry first within each file.memory_recall— Return the markdown content of a topic file, or_index.mdif no topic given.memory_topics— List every topic with entry count, byte size, last-modified.memory_search— Full-text substring search across all topic files (case-insensitive), returns snippets around hits.memory_consolidate— Regenerate_index.mdfrom current topic files.
Use the cb helper to access these as single-line subcommands: cb remember, cb recall, cb topics, cb search, cb learn. See Memory System.
Pixel Streaming Status
A single-source-of-truth tool every agent should call before any /preview capture:
pixel_streaming_status— Returns{module_loaded, signalling_running, streaming_started, ready, viewer_url, viewer_port, streamer_port, preview_url, preview_bare_url, hint}. Wait forready: truebefore capturing to avoid the "Preparing stream / Booting Pixel Streaming" boot overlay.
The cb helper exposes this as cb stream.
13 Editor Event Types
| Event type | Fires when |
|---|---|
selection_changed | Selected actor(s) changed in editor |
actor_spawned | New actor placed in the level |
actor_deleted | Actor removed from the level |
level_changed | Map / sublevel opened or closed |
asset_saved | Any package saved to disk |
asset_imported | New asset finished importing |
asset_added / asset_removed / asset_renamed | Content Browser registry change |
blueprint_compiled | Blueprint just finished compiling |
pie_started / pie_stopped | Play-In-Editor lifecycle |
camera_changed | Editor camera moved (debounced) |
agent_message | Custom message published via publish_agent_event |
Multi-Viewer / Auto-Connect WebRTC
/preview passes AutoConnect=true&AutoPlayVideo=true&StartVideoMuted=true&HoveringMouse=true to the embedded Pixel Streaming 2 viewer. The peer connection establishes on page load with no user gesture. PS2's signaling default MaxPlayerCount=-1 allows multiple humans/agents to view the same editor concurrently.
/preview?bare=1 strips all chrome (header, padding, frame border, HUD chips) so the iframe fills 100vw × 100vh and the stream renders at native browser-viewport resolution — ideal for visual watcher sub-agents.
How to Discover Tools
From Your AI Client
Ask: "What tools are available for materials?" — The AI will call tools/list and filter relevant tools.
From the Dashboard
Click Managers (47) in the ClaudusBridge panel to see all categories with their tool counts.
Programmatically
The auto-generated cb helper script is the preferred path (no JSON-RPC envelope or python -c unwrap):
python Saved/ClaudusBridge/cb.py topics # memory topics
python Saved/ClaudusBridge/cb.py call get_current_map
python Saved/ClaudusBridge/cb.py spawn Cube label=Foo location=[0,0,200] color=red
See The cb Helper CLI for the full subcommand reference.
For raw tools/list, use curl:
curl -X POST http://localhost:3000/mcp \
-H "Content-Type: application/json" \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'
ClaudusBridge Guide Tool
Call ClaudusBridge_guide to get a categorized reference of all tools with descriptions.
Auto-Generated CLAUDE.md
Every MCP initialize regenerates Saved/ClaudusBridge/CLAUDE.md with the full tool reference, organized by category, and threaded with the live tool count from Router::GetTotalToolCount(). Use it as the offline canonical source.
Tool Naming Convention
Tools follow a consistent naming pattern:
| Pattern | Examples |
|---|---|
list_* | list_actors, list_materials, list_blueprints |
get_* | get_actor_details, get_material_graph |
set_* | set_actor_property, set_light_properties |
create_* | create_material, create_variable |
delete_* | delete_actor, delete_node |
add_* | add_node, add_component, add_niagara_emitter |
connect_* | connect_pins, connect_material_expressions |
ai_* | ai_perceive, ai_navigate, ai_engineer, ai_brain_map |
subscribe_* / poll_* / publish_* | Editor events + agent_message bus |
Detailed Reference
For parameter-level detail on every tool, use the ClaudusBridge_guide tool from your AI client, or check the JSON schema files in Resources/tool_schemas/ within the plugin folder. The full machine-readable reference is also written to Saved/ClaudusBridge/CLAUDE.md on every MCP initialize.