What it is
A bridge between an MCP-compatible AI client and Unreal Editor. The assistant can request editor actions through documented tool contracts instead of guessing from plain text.
The professional MCP bridge for Unreal Editor. 1170 registered tools, 1162 callable public contracts across 146 domains, enforced security posture, packaged-plugin distribution, and a validated 1.0 release baseline.
MCP (Model Context Protocol) is a standard way for AI systems to connect to real tools and applications. Instead of only generating text, an AI can use MCP to trigger actions in software like Unreal Engine — and MCP4Unreal is the bridge that makes that connection practical for Unreal Editor.
MCP4Unreal is for developers, technical artists, and studios that want AI-assisted Unreal workflows without dumping the whole project into a chat window.
A bridge between an MCP-compatible AI client and Unreal Editor. The assistant can request editor actions through documented tool contracts instead of guessing from plain text.
Generic setups load too much Unreal context, cost more tokens, respond slower, and leave more room for the wrong tool. MCP4Unreal narrows each request to the domain that matters.
Create or adjust actors, inspect Blueprints, work with materials, trigger Sequencer tasks, query diagnostics, or route other editor operations through validated commands.
Initial 1.0 release for Unreal Engine 5.7 on Windows x64.
Live /health evidence captured on 2026-06-18 from plugin 1.0.0 and engine 5.7.0.
1162/1162 callable public non-alias contracts across 146 domains; 1170 registered tools including 8 compatibility aliases.
Default discovery excludes compatibility aliases and experimental-only entries unless requested.
Public contracts are checked through direct and route validation. The current 1.0.0 evidence records 1162/1162 callable coverage with 1534 fixtures, 1078 typed input contracts, and 0 undeclared required parameters.
Every response declares execution_scoped, handler_owned, or not_applicable — so you always know if a call can undo.
Issues and the roadmap are public on GitHub. Feature suggestions and bug reports are welcome. The plugin is distributed as a packaged build — no source delivery.
A generic MCP setup loads broad Unreal context before the model can act. MCP4Unreal routes only the context a task needs, so far fewer tokens reach the model before execution.
Water, Physics, UI, Audio, Animation, PCG, assets — loaded before execution. Higher cost, slower results, more room for the wrong tool.
A water request loads water context; a UI request loads UI context. Lower cost, faster execution, cleaner Unreal actions.
Estimated token impact per request, compared with broad tools/list loading. The scoped-context routing is validated; the exact reduction depends on the task and client.
MCP4Unreal sits between the MCP client and Unreal Editor. Its job is to keep the model focused before a request becomes an editor operation.
A developer asks for a change in natural language from a terminal, chat client, or IDE.
The client sends the request through the MCP protocol instead of treating Unreal as plain text context.
The model interprets the instruction and selects the relevant Unreal intent.
The bridge loads task-specific context and calls the validated Unreal tool contract.
Unreal receives a cleaner, narrower action that can be audited and validated.
Broad Unreal context reaches the model before the action is selected.
Only the required Unreal context reaches the model before execution.
Current claims are scoped to clients we have actually validated.
Validated live against the running editor on 2026-06-18 over Streamable HTTP with custom headers.
Streamable HTTP and local STDIO validated on 2026-04-28.
Connects directly over Streamable HTTP with custom headers, or through the bundled STDIO bridge. No per-client gateway and no dotted-name rewriting.
Unsupported today: it filters standard dotted MCP tool names such as actors.spawn.
A verified surface across the entire editor.
Spawn, transform, stream, partition, query.
Graph edits, widget trees, skeletal rigs.
Parameter collections, master materials, MI.
Tracks, sections, render queue submissions.
Systems, emitters, cues, chaos simulations.
Git, Git LFS, cook, package, profile.