# Unreal Engine MCP server > Connect Claude, Cursor, or any MCP client to GraphDeck and the AI assistant you already use can build, validate, explain, and deploy Unreal Engine Blueprints and Materials - on a real node canvas, with you approving every change. GraphDeck ships a built-in MCP server (localhost only, off by default) in both the Free and Pro editions. The assistant works on GraphDeck's canvas rather than inside your Unreal project, so mistakes land on a staging ground; with the free Connector plugin it can also deploy into Unreal and compile-check the result. ## An MCP server made for Unreal Engine graphs Connect Claude, Cursor, or any MCP client to GraphDeck and your AI assistant gets a real node canvas to work on: it builds Unreal Engine Blueprints and Materials live while you watch, validates its own output, teaches you how graphs work, and deploys to your project only with your approval. - Free and Pro - Claude Code · Claude Desktop · Cursor · VS Code · any MCP client - Localhost only Links: [Set it up in a minute](https://graphdeck.app/docs/mcp) · Download GraphDeck Free: https://downloads.graphdeck.app/pro ## Pick a path, watch it build The page carries an animated demo of four ways to bring your own AI to GraphDeck, shown for Blueprint, Material and Widget modes. In the MCP path the client connects to the local server and reports the connection like this: ``` graphdeck localhost:30303 · 19 tools ``` and then calls GraphDeck's tools as it works: ``` build_graph GraphDeck validate_graph 97 / 100 · deployable ``` Four ways to bring your AI - the MCP path shows Claude driving the canvas over the built-in server. ### Let your AI assistant use GraphDeck (Best way) MCP server, built in. Connect an assistant like Claude and it drives the whole workflow: it builds graphs live on your canvas, looks things up in your library, and checks its own work against GraphDeck's validation. With the Connector plugin it reaches into Unreal too - pull Blueprints and Materials in, deploy back, compile and verify. You approve anything that touches your library or your project, and Ctrl+Z undoes any build. Claude Code · Claude Desktop · Cursor · VS Code · any MCP client · Free and Pro. See [the MCP setup docs](https://graphdeck.app/docs/mcp). ### Copy-prompt workflow No API key needed. GraphDeck builds a detailed prompt for you. Copy it, paste into whatever AI you already use (ChatGPT, Claude, Gemini), and paste the response back - GraphDeck turns it into a wired, validated graph. Works with any AI · Requires a free account with the AI provider. ### GraphDeck Gems No API key needed. Pre-built Gemini Gems: custom AIs that already know every node, pin type, and syntax rule for each mode. Copy a short prompt, paste it in, copy the response back. Same clipboard flow, better results. Requires a free Google account · Powered by Gemini. - [Blueprints Gem](https://gemini.google.com/gem/1AZR-EdapkR5sVErhXDew4Ky6FG4BETQk?usp=sharing) - [Materials Gem](https://gemini.google.com/gem/1apZZjpVQd4MecDK_VssPmJeKLmPpsQ0c?usp=sharing) ### Direct API key Optional. Plug in your own API key from Anthropic or OpenAI and generate directly inside GraphDeck. Faster if you're doing a lot of generation, and the cost per graph is fractions of a penny. Your key · Your account · Stored locally · Pay per use. ## A canvas, not a text dump Most AI-to-Unreal setups make the assistant describe nodes in text, or poke at your project files directly. GraphDeck's MCP server gives it a proper staging ground instead. ### Build graphs live The assistant generates Blueprints, Materials, and Widgets straight onto the canvas while you watch. Every build is one Ctrl+Z away from gone. ### Check its own work It verifies node names against the real UE node library before using them, and runs every build through GraphDeck's validation pipeline: wrong pins and broken wiring get caught and fixed before you ever see them. ### Teach on the canvas Ask it to explain a graph and it draws the lesson: rings, arrows, and step-by-step walkthroughs anchored to the actual nodes, with the camera gliding between them. ### Use your library It can search your saved graphs and reference material, open files, and build on what you've already made instead of starting cold. ### Deploy to Unreal With the free Connector plugin, the assistant can send a finished graph into your project as a compiled asset - and then compile-check it in the editor to prove it works. ### Stay approved Anything that touches your library or your Unreal project asks first. The assistant proposes; you approve. ## A staging ground, not a scalpel ### AI mistakes land on a canvas, not in your .uassets MCP servers that live inside the Unreal editor hand the assistant your actual project. That's powerful, and sometimes exactly what you want - but it means every AI mistake happens in the real thing. GraphDeck keeps generation, iteration, and teaching on a separate canvas. Your project only changes at one explicit moment: when you deploy. ### It works with Unreal's own MCP, not against it Unreal Engine 5.8 ships a built-in MCP server, and the GraphDeck Connector plugin registers its tools with it (with its own local endpoint on older engine versions). That gives an assistant the full round trip: build and validate in GraphDeck, deploy, then compile and verify inside Unreal. ### Localhost only The MCP server runs on your machine and talks to clients locally. No cloud relay, no account, no telemetry from the MCP surface. ## Connected in about a minute 1. **Enable the MCP server in GraphDeck** - turn it on under Settings → MCP Server. GraphDeck shows the exact local address your client needs, and you pick which capability groups the assistant gets. 2. **Add it to your MCP client** - Claude Code, Claude Desktop, Cursor, VS Code, or anything else that speaks MCP. The [docs](https://graphdeck.app/docs/mcp) have copy-paste config for each. 3. **Ask for a graph** - "Build me a door that opens on overlap." The assistant builds it on the canvas, validates it, and waits for your verdict. ## Common questions ### What is an MCP server for Unreal Engine? MCP (Model Context Protocol) is the open standard AI assistants use to control other software through tools. GraphDeck ships a built-in MCP server, so an assistant like Claude can build, validate, explain, and deploy Unreal Engine Blueprints and Materials through GraphDeck's canvas. ### Which clients work? Any MCP client: Claude Code, Claude Desktop, Cursor, VS Code, and others. If it speaks MCP, it can drive GraphDeck. ### Can the AI break my Unreal project? The AI works on GraphDeck's canvas, outside your project. Deploys are explicit and approved by you, canvas builds undo with Ctrl+Z, and deployed assets are standard Unreal assets you can open, inspect, and compile like anything built by hand. ### Is it free? Yes. The MCP server ships in both GraphDeck Free and Pro, and the Connector plugin is free. ### Do I need the Connector plugin? No. Without it, the assistant builds and validates on the canvas and you paste the result into Unreal via native clipboard export. The Connector adds one-click deploy and the in-editor verification tools. ## Give your assistant a canvas - [Read the setup docs](https://graphdeck.app/docs/mcp) - Download GraphDeck Free: https://downloads.graphdeck.app/pro Unreal Engine is a trademark of Epic Games, Inc. GraphDeck is not affiliated with or endorsed by Epic Games. · © 2026 GraphDeck · Operated by Adam Broder --- [Home](https://graphdeck.app/) · [Docs](https://graphdeck.app/docs/) · [Pricing](https://graphdeck.app/pricing) · [FAQ](https://graphdeck.app/faq) · Contact: hello@graphdeck.co.uk