Unreal MCP Server

The Unreal MCP Server is a Model Context Protocol (MCP) server plugin for Unreal Engine 5.7 that exposes 305 editor tools, 15 resources, and 12 prompts to AI agents via JSON-RPC 2.0 over HTTP. Built by StraySpark.

Feature Overview

Feature	Details
Tools	305 tools across 42+ configurable categories
Resources	15 read-only data sources for project/level/editor context
Prompts	12 reusable workflow templates
Protocol	MCP spec 2025-06-18 (Streamable HTTP) + backward-compatible SSE
Transport	JSON-RPC 2.0 over HTTP on configurable port (default 13579)
Engine	Unreal Engine 5.7, Editor module (Win64, Mac, Linux)
Threading	All UE API calls dispatched to game thread automatically
Undo	All mutating tools wrapped in editor transactions (Ctrl+Z)
Presets	5 tool presets: Full, Scene Building, Gameplay, Minimal, Custom
Status Bar	Live server status indicator in the editor toolbar

How It Works

AI clients such as Claude Code, Cursor, or any MCP-compatible agent connect to the plugin over HTTP. The plugin translates JSON-RPC 2.0 requests into Unreal Editor API calls, executing them on the game thread with full undo support.

AI Client (Claude Code, Cursor, etc.)
    |
    |  HTTP POST/GET/DELETE  (JSON-RPC 2.0)
    v
+--------------------------------------------------+
|  FMCPHttpServer (Streamable HTTP + Legacy SSE)   |
|  Port 13579 (configurable)                       |
+--------------------------------------------------+
|  JSON-RPC Dispatch                               |
|  +-- initialize / ping                           |
|  +-- tools/list -> FMCPToolRegistry (305 tools)  |
|  +-- tools/call -> Tool Handler (game thread)    |
|  +-- resources/list/read -> FMCPResourceProvider |
|  +-- prompts/list/get -> FMCPPromptProvider      |
+--------------------------------------------------+
|  UMCPSettings (presets, category toggles,        |
|  rate limiting, security, verbose logging)        |
+--------------------------------------------------+

Each tool handler executes on the editor's game thread via AsyncTask(ENamedThreads::GameThread, ...). All mutating operations are wrapped in GEditor->BeginTransaction() / EndTransaction() for full undo support.

Tool Categories at a Glance

The 305 tools are organized into 42+ categories grouped by domain:

• Core -- Actor (14), Editor (7), Asset (6), Level (6)
• Scene Building -- Material (5), MaterialGraph (8), StaticMesh (7), Batch (3), Environment (4)
• Scene -- Spatial (10)
• Scripting -- Blueprint (51), Python (1), Search (2)
• UI -- Widget/UMG (16), CommonUI (4)
• AI Generation -- AI Image (2), AI 3D Model (3)
• Animation -- Animation (5), AnimGraph (14), ControlRig (2)
• Cinematic -- Sequencer (12)
• Gameplay -- GAS (8), EnhancedInput (6), GameFramework (6), GameplayTags (3)
• AI -- AI/BehaviorTree (8), StateTree (5)
• Physics -- Physics (9), Navigation (3)
• World -- Spline (7), WorldPartition (2), Landscape (3), Foliage (4)
• VFX and Audio -- Niagara (3), Audio (3), MetaSound (6)
• Data -- DataTable (3), DataStructures (3)
• Networking -- Networking (5)
• Workflow -- Macro (6), Build (7), EngineAPI (3), AssetManagement (7), Performance (4)

See the full Tool Categories reference for details on every tool.

Presets

Choose a preset to control which tool categories are available and manage AI context token costs:

Preset	Tools	Use Case
Full	~305	All 42+ categories enabled. Maximum capability.
Scene Building	~170	Core + Scene + Scripting + Cinematic + World + VFX + Workflow.
Gameplay	~190	Core + Scripting + Gameplay + AI + Data + Physics + Workflow.
Minimal	~28	Actor + Editor + Level only. Lowest context cost.
Custom	Variable	Per-category toggles for fine-grained control.

Resources and Prompts

The plugin provides 15 read-only resources for context awareness (project info, level data, performance stats) and 12 reusable prompts for common workflows (world building, blueprint development, debugging). The server uses schema-on-demand context optimization, delivering tool schemas only when needed to minimize AI context token costs.

Next Steps

• Getting Started -- Install the plugin and connect your first AI client
• Installation -- Detailed installation, configuration, and troubleshooting
• Architecture -- Deep dive into the plugin's internal design
• STDIO Bridge -- Connect clients that only support STDIO transport
• API Reference -- Core classes, settings, HTTP endpoints, and tool registration