Beat Claude runs through the Senserva MCP inside Claude, against a built-in simulated Microsoft 365 tenant. No registration, no key, no access to your own tenant.
Grab the signed build for your OS. One small binary, no installer, no agents.
Run demo claude. It builds the demo database and automatically wires the Senserva MCP into Claude, no config to edit.
Install the free Claude Desktop app if you do not have it; the MCP config is already in place. Open Claude and say “want to play a game” Full MCP setup.
Illustrative result from the scripted Claude-versus-Alex match below. Your scores are yours.
Game Mode turns your Microsoft 365 security data into a live, graded match with Claude, across five skill levels on real findings. Spot something Claude gets wrong and you steal bonus points. Get his answers correct and you get points too.
MS.AAD.3.1v1, MCSB IM-6, bridged 800-171 IA-2(1). MS.AAD.5.1 (OAuth scope governance), bridged 800-171 AC-6(7). MS.INTUNE baseline + MS.AAD.3.1v1, bridged 800-171 CM-6 + AC-2(7). Most security training tests whether you remember a definition. You v. Claude tests whether you can read a real configuration and say what is wrong, why it matters, and what an attacker would do with it. Every challenge is drawn from a live Senserva scan of the demo tenant, so you practice on the same Microsoft 365, Intune, Defender, Entra ID, and Purview findings the product surfaces in production.
The questions climb five skill levels, from MFA basics to multi-domain attack chains, and the twist that makes it stick: when you catch Claude getting an answer wrong, you score the most points of all. That single mechanic forces you to reason from evidence instead of agreeing with the confident-sounding AI.
Here are real questions from the game, and what a player walks away understanding. These are the same patterns that close audit findings in the real world.
The question: A Global Administrator at the demo tenant has no MFA registered. What is the worst that could happen?
What you learn: Why privilege concentration turns a small gap into a total-takeover event. Global Administrator holds every role across every Microsoft 365 workload, so one leaked password equals full tenant compromise, including all data. You learn to rank a finding by who the account is, not just what the setting is, and to map it to SCuBA MS.AAD.3.1v1 and MCSB IM-6 for an auditor.
The question: A marketing app moved from Files.Read.Selected to Files.ReadWrite.All in the last 30 days. How much did the blast radius change?
What you learn: The difference between scoped and tenant-wide permissions, which most people misjudge. Files.Read.Selected is a curated handful of files; Files.ReadWrite.All is every drive in the tenant with write access. That is a categorical jump, not a small one, and it happened in a single consent click. You learn to read OAuth scopes precisely, and this is a question where players routinely catch Claude conflating Selected with All.
The question: A re-enabled privileged account signs in from a BYOD device that last checked in as Compliant 47 days ago, the same day the account was disabled. The CA policy requires a compliant device. What is the attack chain?
What you learn: How controls that each look healthy combine into a hole. The device record froze in a Compliant state when check-ins stopped; Conditional Access still trusts that stale state, so the gate passes for an attacker who has the credentials and a privileged role waiting. The policy is not broken, the staleness window is. You learn to reason across identity, device, and policy together, the kind of multi-domain chain only a graph correlates.
The question: Two devices each have a different missing patch. One CVE is in the CISA KEV catalog with a high EPSS score; the other has a higher CVSS base score but no known exploitation. Which do you deploy first?
What you learn: Why a CVSS number alone is the wrong way to prioritize. Known exploitation (CISA KEV) and exploitation probability (EPSS) beat raw severity, because a Critical that nobody is exploiting is less urgent than a High that ransomware is using today. You learn the defensible triage order Senserva uses: actively exploited first, then severity, then exposure age and fleet impact.
The question: A sensitivity label is published but applied to almost nothing, and a Copilot-connected app has broad read access. Where is the data-governance risk?
What you learn: That a labeling policy is only as good as its coverage, and that AI assistants inherit whatever oversharing already exists. You learn to connect Purview label coverage, unified audit log health, and Copilot and AI agent permissions into one data-exposure picture, the governance side of Microsoft 365 security most quizzes ignore.
We use Google Analytics cookies to understand site traffic. No findings, scan data, or tenant data are sent. Privacy policy.