This is the department the Unreal MCP was practically built for. You describe intent in plain English — "scatter cover through this courtyard so no sightline runs longer than 15 meters" — and Claude executes it in the live editor. Blockouts, balance passes, Blueprint prototypes, and design docs all move at the speed of a sentence.
Prototype-grade Blueprints, scene tweaks, and data changes no longer queue behind engineering. You direct; Claude builds; engineering hardens what survives playtest.
When a layout variation costs one sentence instead of one afternoon, you try eight ideas instead of two. More reps is the whole job — Claude buys you reps.
Balance data stops being a wall of numbers. Ask what happens to time-to-kill if armor scales differently, and get the curve, the outliers, and the edited table back.
You don't need to write a line of code to use any of these. Talk to Claude the way you'd brief a very fast level scripter who never mishears you.
Browse all Design skills in the Library →
The workhorse. Through the Unreal MCP, Claude spawns, moves, and arranges actors, edits properties in bulk, queries the scene ("list every enemy spawn point in this map and its patrol route"), and runs console commands — with our project's folder rules and actor conventions baked in.
Blockouts, set-dressing passes, mass tweaks, spawn-point audits: things that used to be an afternoon of clicking are now a paragraph of intent.
Claude reads and builds Blueprint graphs through the MCP, following our conventions so your prototypes are legible to engineering instead of spaghetti. It also knows our BP-vs-C++ boundary — so when your prototype proves out, the handoff to native code is a request, not a rewrite negotiation.
After any change, Claude launches PIE, exercises the thing it just built, watches the log, and screenshots the result. You start your playtest from "it runs" instead of "does it run?"
Our data tables and curve assets, read and edited safely. Claude models the downstream effects of a tuning change before making it — damage curves, economy flow, progression pacing — then applies it across every affected row instead of you hand-editing forty cells.
Checks your levels and content against studio rules — naming, folder placement, missing references, actors that will break the cook. Run it before you submit and your changelists stop bouncing.
Talk through a mechanic for ten minutes; get a structured one-pager back. Better: after the mechanic changes in-engine, ask Claude to diff the doc against what's actually built and update it. Docs stop lying.
First-pass ambient dialogue, item descriptions, and lore variants in the game's established voice — generated in batches, curated by you. You stay the author; Claude is the writers' room that never runs dry.
"How do the best roguelites handle meta-progression pacing? Give me the patterns, the outliers, and what would fit HEX." A structured competitive teardown in minutes, not a weekend of YouTube.
Dump raw playtest notes, survey exports, and session recordings' transcripts on Claude; get back themes, severity, and the three changes most likely to move the fun. Signal from noise, same day.
If you find yourself correcting Claude twice about the same thing — metrics, naming, "we never place ammo in boss arenas" — that's not a prompt fix, that's a skill update. Say: "add this rule to editor-ops and open a PR." Every designer's Claude learns it at once. See exactly how →
Same idea as the Unreal MCP on the Setup page — a live wire between Claude and a tool you use, instead of a written playbook. Every command below is verified against each product's own current docs, not guessed. Add one with -s user so it's yours, not the team's — same rule as the Setup page.
If your GDDs live in Notion, Claude reads and updates them directly — no more pasting context in every session.
Then type /mcp and follow the browser sign-in. First time only: open each page you want visible → ••• → Add connections → select Claude.
UI flow wireframes, HUD and menu mockups. Claude reads a file's actual structure, not just a screenshot of it.
Then /mcp → sign in with your Figma account. Needs Dev Mode enabled on the file.
Progression-system flowcharts, quest state diagrams, retro boards — read and edited, not just described.
Then /mcp → sign in. Enterprise-plan boards may need an admin to enable MCP access first.
If design tasks live in Linear instead of our production tracker, same benefit as Notion — Claude reads and updates issues directly.
Then /mcp → sign in. No OAuth on your account? Use a Personal API key instead (Linear → Settings → Security & access).
Real-time web research with sourcing — competitive teardowns, "how do the top roguelites handle X," in one prompt.
Get a key at perplexity.ai/account/api/group.
A second, independent search engine — useful when you want results that don't route through one company's index.
Get a free key (2,000 queries/month) at brave.com/search/api.
Also useful for comp research, no packaged MCP yet: IGDB (structured metadata on thousands of games) and the Steam Web API (player counts, review sentiment). Solid APIs — ask engineering to wrap one if it's worth the time.
TTS for barks, item descriptions, and dialogue read-throughs before a VO budget exists. Hear the line before you commit to it.
Get a key at elevenlabs.io/app/settings/api-keys (free tier: 10k credits/month).
For designers who blockout in Blender before handing off to art. The dominant community MCP project — actively maintained, tens of thousands of users.
Two parts. In Blender: install the addon from github.com/ahujasid/blender-mcp (Preferences → Add-ons → Install), enable it, start its connection server from Blender's sidebar. Then on the Claude side:
No API key needed for core use.
Relational content databases — item tables, quest lists, dialogue trees — with real structure instead of a flat spreadsheet.
Then /mcp → sign in. Alternative: a Personal Access Token from Airtable → Account → Developer hub.
Google's official Sheets MCP is real, but it's still in Developer Preview — it needs its own Google Cloud project and OAuth setup, not a one-line install. If balance-in-Sheets matters to your workflow, ask engineering to provision one shared connector for the team rather than setting it up yourself.