Architect + Claude MCP = Magic (Part 2)

Earlier this year I wrote about Rogue Architect and teased something bigger coming. This is that post.

So buckle up… you won’t ever build labs the same way again :)

When we first floated the idea of typing "Build me a large Star Wars-inspired Enterprise Red Team scenario" and watching an entire lab appear, honestly, we weren't sure how far we could push it.

We knew the canvas was solid. We knew the plugin library was vast. But the intricacies of building large-scale complex Active Directory scenarios bring on a lot of complexity.

Could AI actually drive all of it? Would it actually build without a ton of extra final configuration by a user?

Well, we've spent the last few months wiring Rogue Architect into Claude through MCP, and I can say without hesitation:

It's magic. It's also completely changed how we build Red/Blue labs here at Rogue Labs.

What we ended up with is four Claude Code skills sitting on top of the Rogue Arena MCP toolkit (100+ MCP tool calls). Each one takes a slice of lab work that used to be hours of clicking, researching, or writing Ansible, and turns it into a conversation.

Now, let's take a tour of our 4 new Skills and look at how you can build HUGE, hyper-realistic Active Directory scenarios.

1. /architect-brainstorm

This is the one we teased in Part 1. Type a prompt like "Build me a mid-size tech startup red team scenario with a parent domain, a subsidiary domain, and a DMZ," and Claude immediately flips into brainstorm mode, asking you questions and walking you through four distinct phases of scenario creation.

Phase 1: The Shape. Claude asks a few questions about the company you want to build on the canvas. Name, industry, size, security maturity. Then it drafts the VLANs, machines, domains, and high-level infrastructure.

Phase 2: The Cast. Claude dispatches parallel subagents to generate characters for every Domain, and even assigns them workstations from your scenario. Names, titles, hobbies, password habits, file organization habits, relationship events between each other, who reuses credentials, who hates IT. Real people, your red team will actually recognize when they pop a box and read someone's Outlook. Building a Red Rising-themed scenario? Key players from fictional universes are assigned roles in the company that just make sense (shoutout to Darrow, the CEO).

Phase 3: The Build. With the high-level plan mapped out, Claude immediately sets to work querying the Rogue Architect plugin repository, mapping plugins to every machine. Domain controllers get “Create DC”. Workstations get “Join to Domain”. Users are set to auto-login. And the fun part, Claude dynamically generates hundreds of files to “seed” across systems in the scenario. These files match the user insights built in Phase 2 above, and their role at the company. Don’t be surprised to see a letter from HR sitting on the IT Admins Desktop repremanding them for watching Netflix during their shift :)

Phase 4: The Heist. (Optional.) With your direction and insight, Claude builds a full exploit chain across your scenario and optionally builds a “crown jewel” target at the end of the chain. Initial access, multi-hop pivots, vulnerability chains. If you want Kerberoast into ADCS into Tier 0, just ask.

All phases are completely relaxed…you converse with Claude to update plans, make tweaks to exploit paths, etc. This ensures you end up with the exact scenario you were hoping for.

Phase 5: Implementation.

The phase is the best of all. With the rough plan in place, Claude tells you to go grab some coffee and dynamically builds out your entire multi-domain Active Directory scenario in Rogue Architect via the MCP tooling. Every system, plugin, and user is accounted for. VLANs, machines, plugins, parameters, exploit paths, all of it lands as draft changes. Once complete, you review the draft changes, click accept to apply them to your scenario, and click build to provision the scenario into VMs that you can log in to and control.

And on the off chance the deployment fails, Claude has an active debugger skill and can query YML Ansible logs, check plugin source code, and pinpoint any failures.

This one alone has been worth the entire project. Scenarios that used to take a full week of careful design now happen in 15 minutes of back-and-forth.

And the best part? Once the scenario is built. Every one of the 100+ MCP tools is still available to Claude. Just run /architect-freeform and command Claude to investigate your current scenario, check on users present, add one-off files, and way more. Iterative building is still just as easy.

2. /rogue-curriculum-builder

Here's the thing about a great lab. Curriculum can MAKE or BREAK the student experience.

Nobody wants to dig through thousands of lines of wiki walkthroughs. An interactive CLEAN curriculum should accompany it. A 50-machine Active Directory forest is just a pile of VMs without an engaging story, objectives, or a learning path to go with it.

Our Claude curriculum plugin turns this all into a conversation, too.

With your scenario built, just tell Claude how you want to break up the CTF/Scenario tasks. Explain the section/chapter layout, what content should go in each chapter, etc.

Building a “lunch and learn” for your team on ADCS?

Tell Claude to design an epic ADCS scenario based on a real-world e-commerce enterprise network. Then tell Claude what you want the lab curriculum to look like.

Everything is done and ready to test. Chapters, sections, rich text, embedded videos, code blocks, multiple choice questions, file upload questions, flag gated progression, unlock keys, all of it.

It handles both LINEAR layouts (step-by-step courses) and CTF graph layouts (nodes with edges where students pick their path). It also bulk populates efficiently, so when you ask for "twelve sections covering Windows persistence," you don't sit there watching it build one at a time.

Internally, I've started outlining what I want to teach, handing it to Claude, and iterating on the output instead of staring at an empty chapter editor. It's a completely different drafting experience.

3. /rogue-active-deployment

Quick aside on this skill…it doesn’t have anything to do with Rogue Architect, but its a game changer for prepping for Red Team operations.

Here's the headline: automated red team TTP development is here.

Learn more at our sister post HERE

4. /rogue-plugin-dev

Part 1 mentioned our plugin library has over 100 plugins and that we're always adding more. The truth is, plugins are a ton of work.

Writing Ansible that installs offline, pulling the right artifacts, wiring up parameters, uploading vault files, and testing end-to-end.

Well… plugin dev just became a conversation, too.

Say you want a plugin that installs a specific Defensive agent to test against. Claude researches the offline install path (apt, rpm, MSI, Chocolatey, Docker image, whatever fits), decides whether it should be a single plugin or a set, scaffolds the project, writes the metadata, drafts parameters with realistic sample values, and generates download scripts for any artifacts.

Then in the development loop, you iterate. Claude rips into a failing deployment via MCP tools, finds bugs, updates the plugin code, changes out any vault files, adjusts user-specified parameters, and then kicks off a fresh build. In a loop. Until it's green.

We've been using this skill internally and most recently built a massive Exchange Email plugin which seeds mailboxes, artifact emails, and other complicated tasks, and it's as easy as a breeze.

Wanna check it out?

I want to be clear about one thing. These plugins aren't a gimmick or a demo. They're how we build labs at Rogue Labs now.

The Star Wars scenario from Part 1? Real. We've built it. We've also built a Dunder Mifflin themed scenario, three pharma companies, a fintech, and a municipal water utility.

Each one with different Active Directory topologies, full character backstories, seeded files, and working exploit paths. Each one started with a sentence and a ten minute conversation.

Part 1 was about making lab building visual and drag and drop. Part 2 is about making it conversational. Same platform. Same canvas. Same plugins. Just an AI co-pilot sitting on top of all of it through MCP.

And we're just getting started.