Editor commands, Python execution, PIE, undo/redo, hot reload, viewport, performance, sequencer, build pipeline, logs, editor control. Actions: - start_editor: Launch Unreal Editor with the current project and reconnect bridge - stop_editor: Close Unreal Editor gracefully - restart_editor: Stop t...
Risk signalsAccepts freeform code/query input (command) · Accepts file system path (filePath) · High parameter count (64 properties) · Bulk/mass operation — affects multiple targets · Admin/system-level operation
Part of the Ue server.
Free to start. No card required.
AI agents may call editor to permanently remove or destroy resources in Ue. Without a policy, an autonomous agent could delete critical data in a loop with no way to undo the damage. PolicyLayer blocks destructive tools by default and requires explicit human approval before enabling them.
Without a policy, an AI agent could call editor in a loop, permanently destroying resources in Ue. There is no undo for destructive operations. PolicyLayer blocks this tool by default and only allows it when a human explicitly approves the action.
Destructive tools permanently remove data. Block by default. Only enable with explicit approval workflows.
{
"version": "1",
"default": "deny",
"hide": [
"editor"
]
}See the full Ue policy for all 22 tools.
These attack patterns abuse exactly the kind of access editor gives an agent. Each links to the full case and the policy that stops it:
Other destructive tools across the catalogue. The same approach applies to each: deny by default, or require human approval.
Editor commands, Python execution, PIE, undo/redo, hot reload, viewport, performance, sequencer, build pipeline, logs, editor control. Actions: - start_editor: Launch Unreal Editor with the current project and reconnect bridge - stop_editor: Close Unreal Editor gracefully - restart_editor: Stop then start the editor - build_project: Build the project's C++ code using Unreal Build Tool. Editor should be stopped first. - execute_command: Run console command. Params: command - execute_python: Run Python in editor. Params: code - run_python_file: Run a Python file from disk with __file__/__name__ populated (#142). Params: filePath, args? - set_property: Set UObject property. Params: objectPath, propertyName, value - play_in_editor: PIE control. Params: pieAction (start|stop|status), waitForAssetRegistry? (start only; default true - block until the AssetRegistry initial scan completes before requesting PIE, otherwise PIE silently no-ops on cold editor starts), assetRegistryTimeoutSeconds? (default 180) (#406) - get_runtime_value: Read PIE actor property. Params: actorLabel, propertyName (supports dotted paths: component.field or component.struct.field for nested reads on component subobjects, #344/#381) - get_pie_pawn: Resolve the controlled pawn in the active PIE world. Params: playerIndex? (default 0). Returns actorLabel/class/location/rotation (#228/#229) - invoke_function: Call a BlueprintCallable / Exec UFUNCTION on a target actor or one of its components. Params: actorLabel, functionName, component? (component subobject name; redirects target from the actor to that component, #382), args? (object), actorArgs? (object mapping UObject* parameter name to actor label, resolved against live actors in the active world; #383), world? (editor|pie). Returns out/return params (#228/#229) - set_pie_time_scale: Fast-forward PIE game time. Params: factor (>0). Raises WorldSettings caps and calls SetGlobalTimeDilation. - hot_reload: Hot reload C++ - undo: Undo last transaction - redo: Redo last transaction - get_perf_stats: Editor performance stats - run_stat: Run stat command. Params: command - set_scalability: Set quality. Params: level - capture_screenshot: Screenshot. Params: filename?, resolution?, target? (auto|pie|editor; auto routes to PIE viewport when PIE is running) (#226) - capture_scene_png: Headless PNG screenshot via SceneCapture2D (works unfocused, guaranteed RGBA8 LDR). Params: outputPath, location?, rotation?, width? (default 1280), height? (default 720), fov? (default 90) (#148) - get_viewport: Get viewport camera - hit_test_viewport_pixel: Ray-cast from a screen pixel through the active editor viewport and return the first hit. Builds the ray from the live viewport's projection matrix (no FOV/aspect guessing). Returns hit + actorLabel/actorClass/componentName/componentClass/materialPath/location/impactPoint/normal/distance/faceIndex/boneName/physicalMaterial. Params: x, y (pixel coords), width? height? (override viewport size when picking from a different-resolution screenshot), maxDistance? (default 200000), ignoreActors? (array of actor labels) (#418) - get_runtime_values: Bulk runtime read across the active world. For each actor/component matching classFilter, resolves every path against the (actor|component) root and returns rows of {actorLabel, actorClass, componentName?, componentClass?, values, errors?}. Paths support property hops, sub-object hops, and zero-arg BlueprintCallable getter calls at any segment (e.g. 'PowerConnector.GetRequired' reaches a UFUNCTION on a UObject sub-object). classFilter matches actor class OR component class - omit to match everything. World defaults to PIE if running, else editor. Params: classFilter?, paths[], world? (editor|pie) (#414) - set_viewport: Set viewport camera. Params: location?, rotation? - focus_on_actor: Focus on actor. Params: actorLabel - create_sequence: Create Level Sequence. Params: name, packagePath? - get_sequence_info: Read sequence. Params: assetPath, includeSectionDetails? (attach sockets, first transform key values per track) - add_sequence_track: Add track. Params: assetPath, trackType, actorLabel? - play_sequence: Play/stop/pause sequence. Params: assetPath, sequenceAction - build_all: Build all (geometry, lighting, paths, HLOD) - build_geometry: Rebuild BSP geometry - build_hlod: Build HLODs - validate_assets: Run data validation. Params: directory? - get_build_status: Get build/map status - cook_content: Cook content. Params: platform? - get_log: Read output log. Params: maxLines?, filter?, category? - search_log: Search log. Params: query - get_message_log: Read message log. Params: logName? - list_crashes: List crash reports - get_crash_info: Get crash details. Params: crashFolder - check_for_crashes: Check for recent crashes - set_dialog_policy: Auto-respond to dialogs matching a pattern. Params: pattern, response - clear_dialog_policy: Clear dialog policies. Params: pattern? - get_dialog_policy: Get current dialog policies - list_dialogs: List active modal dialogs - respond_to_dialog: Click a button on the active modal dialog. Params: buttonIndex?, buttonLabel? - open_asset: Open asset in its editor. Params: assetPath - reload_bridge: Hot-reload Python bridge handlers from disk - save_dirty: Flush every dirty package and return a per-package saved/failed map. Use after multi-step CDO/component edits when set_class_default leaves the asset dirty without persisting (#378). Params: includeMaps? (default true), includeContent? (default true) - configure_pie: Set ULevelEditorPlaySettings - multi-client PIE, net mode, single-process flag. Params: numClients?, netMode? (standalone|listen|client), runUnderOneProcess?, launchSeparateServer? (#384) - get_pie_config: Read current ULevelEditorPlaySettings (numClients, netMode, single-process, separate-server) (#384) - list_dirty_packages: Enumerate currently-dirty content + map packages (#340). It is categorised as a Destructive tool in the Ue MCP Server, which means it can permanently delete or destroy data. Block by default and require explicit approval.
Register the Ue MCP server in PolicyLayer and add a rule for editor: allow, deny, rate-limit, or require approval. Point your MCP client at the PolicyLayer proxy URL and the rule is enforced on every call, before it reaches Ue. Nothing to install.
editor is a Destructive tool with critical risk. Critical-risk tools should be blocked by default and only enabled with explicit human approval.
Yes. Add a rate_limit block to the editor rule in your PolicyLayer policy. For example, setting max: 10 and window: 60 limits the tool to 10 calls per minute. Rate limits are tracked per agent session and reset automatically.
Set action: deny in the PolicyLayer policy for editor. The AI agent will receive a policy violation error and cannot call the tool. You can also include a reason field to explain why the tool is blocked.
editor is provided by the Ue MCP server (ue-mcp). PolicyLayer sits as a proxy in front of this server to enforce policies before tool calls reach the server.
Deterministic rules across all 22 Ue tools. Per-identity grants. Full audit log. Live in minutes. Nothing to install.
Free to start. No card required.
4,600+ MCP servers and 31,000+ tools scanned and risk-classified.