Natfii/UnrealClaude

npm install -g @anthropic-ai/claude-code && claude auth login
git clone --recurse-submodules https://github.com/Natfii/UnrealClaude.git
Build the plugin from source with RunUAT BuildPlugin (no prebuilt binaries included), then copy the output to your project's Plugins/ or engine plugins folder
Install MCP bridge deps: cd <PluginPath>/UnrealClaude/Resources/mcp-bridge && npm install

Usage

• •Open the editor and use Tools > Claude Assistant; the MCP server starts automatically on port 3000
• •Ask questions like 'How do I create a custom Actor Component in C++?' or 'Write a BlueprintCallable function that spawns particles'
• •Query the dynamic UE context with the unreal_get_ue_context tool by category (animation, blueprint, slate, actor, assets, replication) or keywords
• •Optionally launch Claude Code CLI pointed at the UnrealClaude install and restart it after the all-clear prompt

Key features

• ✓Native editor chat panel with live streaming, tool call grouping, and code block rendering; session persistence in Saved/UnrealClaude/
• ✓MCP server with 20+ tools: actor, level, Blueprint, Animation Blueprint, asset, character, material, enhanced input, and utility tools
• ✓Dynamic UE 5.7 context loader with on-demand API documentation
• ✓Script execution (Python / Console / Editor Utility, C++ via Live Coding) gated by permission dialog (deny-by-default)
• ✓Project-aware: gathers modules, plugins, settings, recent assets, custom CLAUDE.md; uses Claude Code auth (no separate API key)

Best for

Unreal Engine 5.7 developers who want AI coding assistance inside the editor with real UE context and Blueprint/level automation.

Caveats

• ⚠Requires Unreal Engine 5.7 and Claude Code CLI with authentication (Claude Pro/Max subscription or API access); must be built from source per platform/engine version
• ⚠Blueprint/Animation Blueprint tools noted as 'few bugs still, don't rely on fully'
• ⚠Scripting and console tools are hidden by default (HIDDEN_TOOL_NAMES); unhiding run_console_command is treated as enabling arbitrary code execution — keep to trusted setups
• ⚠MCP port 3000 can conflict; large projects/global Claude Code plugins (Superpowers, context7) can slow responses up to 3+ minutes
• ⚠Auto-approve script execution setting is off by default and explicitly only for trusted machines/projects