I recently watched something that honestly changed how I think about cloud operations. A colleague connected Claude — Anthropic's AI — directly to our Azure environment via the Azure MCP integration and typed a simple prompt: "Audit the networking in this subscription and tell me what looks wrong."
What came back wasn't a generic checklist. Claude walked through the environment methodically, cross-referencing VNets, peering configurations, route tables, NSGs, and firewall policies — and surfaced a list of real problems that had been silently sitting there for months.
Empty VNets burning cost. Peerings that were half-configured, connected on one side but not the other. And perhaps most concerning: VNets with live workloads that had no route to the hub firewall — meaning traffic was flowing completely uninspected.
None of it was catastrophic on its own. But all of it was invisible until AI looked at it with fresh eyes.
Why cloud networks quietly accumulate debt
Enterprise Azure environments grow fast and rarely shrink cleanly. A project spins up a VNet, gets cancelled, and the network resources stick around. A migration team peers a spoke to the hub but misconfigures the UDR, so traffic routes around the firewall instead of through it. An ops team creates a test environment, links it up, and then... moves on.
No single person has a complete picture of the whole network at any given time. Audits are expensive, infrequent, and usually scoped to specific compliance requirements rather than general hygiene. The result is that most enterprise Azure environments carry a quiet backlog of misconfigurations, orphaned resources, and security gaps that nobody is actively looking for.
The core problem: Manual network audits require a skilled engineer to spend days pulling data from the Azure portal, cross-referencing resource configurations, and building a picture of how traffic actually flows. Most organizations do this once a year at best — if at all.
What Claude actually found
In the audit I witnessed, Claude identified three categories of issues that are extremely common in mature Azure environments:
Isolated VNets bypassing the hub firewall
Spoke VNets with live resources but no UDR forcing traffic through the hub NVA — effectively unprotected from lateral movement.
Empty and orphaned VNets
VNets with no subnets or no deployed resources, some still peered to the hub — consuming address space and adding noise to the topology.
Half-disconnected peerings
Peering relationships configured on one side only, leaving the connection in a broken state that doesn't fail loudly — it just silently doesn't work.
What's notable isn't just that Claude found these things — a skilled network engineer would too, given enough time. What's notable is the speed and completeness. Claude looked across the entire subscription simultaneously, connected the dots between resource types, and explained each finding in plain language that both engineers and managers could act on.
How it works: Claude + Azure MCP
The integration that makes this possible is the Azure Model Context Protocol (MCP) — a standardized interface that allows AI models like Claude to interact with external systems, APIs, and data sources in real time. Rather than copying data into a prompt manually, MCP lets Claude query Azure's Resource Manager APIs directly during the conversation.
How a Claude Azure audit works
Importantly, the MCP integration operates with read-only access by default. Claude can see your environment and reason about it — it doesn't make changes unless you explicitly configure it to do so and approve each action. This makes it safe to run as a regular audit tool without risk of unintended modifications.
AI auditing vs. traditional approaches
| Capability | Manual audit | Scripted tooling | Claude via MCP |
|---|---|---|---|
| Time to complete | Days to weeks | Hours (setup) + minutes (run) | Minutes |
| Finds known issue patterns | Yes | Yes (if scripted) | Yes |
| Finds novel or unexpected issues | Yes (if experienced) | No | Yes |
| Explains findings in plain language | Yes | Rarely | Yes |
| Cross-references multiple resource types | Depends on engineer | Limited | Yes, simultaneously |
| Can answer follow-up questions | Yes | No | Yes |
| Cost | High (engineering time) | Medium (build + maintain) | Low |
This is the direction enterprise cloud ops is heading
The traditional model of cloud operations — scheduled audits, manual reviews, siloed tooling — was designed for a world where infrastructure changed slowly and environments were small enough to reason about manually. Neither of those things is true anymore.
AI models connected to live cloud environments via MCP represent something genuinely different: an always-available, conversational interface to your infrastructure that can reason about it holistically, explain what it finds, and help you prioritize what to fix. Not as a replacement for experienced engineers — but as a force multiplier that lets them do in minutes what used to take days.
The organizations that figure this out first are going to have a meaningful operational advantage. They'll find security gaps before attackers do. They'll catch waste before it compounds. And they'll spend less time staring at the Azure portal and more time actually improving their environments.
The technology is available today. The question is whether your team is using it.
Want to run a Claude audit on your Azure environment?
We can help you set up the Azure MCP integration, scope the audit, and interpret the findings. Get in touch and let's talk through what your environment might be hiding.
info@acendri-solutions.com