· via dev.to (home feed)
Microsoft Defender for Linux: dev.to guides cover install, benchmarks and licensing
Two dev.to guides walk through installing Microsoft's mdatp agent on Ubuntu, Fedora and Arch, claiming low overhead and container support — but they disagree on licensing and benchmark numbers.

What Defender for Linux is, according to the guides
Two guides published on dev.to on 19 October 2026 — one in English, one in Spanish, both by the same author — walk system administrators through deploying Microsoft Defender for Linux, the mdatp agent, on Ubuntu 26.04, Fedora 40 and Arch Linux. Both describe it as a full endpoint protection platform rather than a plain antivirus: a hybrid engine pairing local signatures with cloud AI scoring for zero-day detection, EDR telemetry feeding the Microsoft Defender for Endpoint console, built-in quarantine with policy-driven remediation, and a unified view across Windows, macOS, Linux and Android endpoints.
Installation across Ubuntu, Fedora and Arch
Both guides route installation through Microsoft's package repository. On Ubuntu 26.04, you import Microsoft's GPG key, add the prod repository and install mdatp with apt. On Fedora 40, the same key is imported into rpm and the agent is installed with dnf from a repository the English guide configures against Microsoft's rhel/9 path. Arch users install from the AUR, though the posts disagree on the package name — mdatp via a helper such as yay in one, microsoft-defender-atp built with makepkg in the other.
They also differ on activation. The English guide enables real-time protection with mdatp health --set real_time_protection=on, while the Spanish one onboards machines to a cloud tenant via mdatp onboarding --cloud-tenant. Both finish by checking agent status with mdatp health.
Competing benchmark numbers
The English guide reports average CPU overhead of 2.3 percent on an idle server, 3.8 percent while compiling, 4.5 percent during Docker builds and 5.1 percent on a file server handling 10,000 operations per second, with idle memory use of roughly 150 MiB. It says the figures were gathered on a 12-core Intel Xeon E5-2670 v3 with SSD storage and are comparable to or better than ClamAV and Sophos.
The Spanish guide gives different numbers: 3 percent CPU during a full scan on an Intel i7-12700 and 120 MB of RAM at idle. Its comparison table has Defender detecting 96 percent of what the article describes as a ransomware suite generated by GPT-4, ahead of Bitdefender at 92 percent, Sophos at 89 percent and ClamAV at 71 percent. The idle memory figures and competitor sets do not match between the two posts, and neither publishes a methodology beyond brief hardware notes.
Automation and containers
Both guides include Ansible roles that add the Microsoft repository, install mdatp and activate protection or onboarding, so new nodes are protected automatically. Their Terraform examples diverge: the English one provisions an Azure Linux virtual machine and injects the install script through cloud-init, while the Spanish one applies a Microsoft.AzureDefender extension to an AKS cluster. The Spanish guide also claims the agent runs without systemd — inside Docker containers or on OpenRC-based systems — by setting MDE_DISABLE_SYSTEMD=1, and shows an Alpine Dockerfile that runs the mdatp daemon directly.
Use cases highlighted across both posts include WSL2 development environments, CI/CD runners that scan container images before push, and data-science workstations where the agent is said to guard Jupyter kernels and Python dependencies. The English guide adds edge devices, citing a small memory footprint and an offline mode for air-gapped networks.
Privacy and licensing
Both guides claim telemetry is limited to metadata — file hashes, paths and threat types — encrypted with TLS 1.3, with no file contents or personal data transmitted. Both describe ways to switch cloud reporting off and forward logs to an internal SIEM over syslog, though the exact commands differ between the posts. The Spanish article adds a default 30-day retention period configurable to internal policy, and the English one lists certifications including ISO 27001 and SOC 2.
Licensing is where the sources flatly disagree. The English guide says enterprise licensing is tied to Microsoft 365 and E5 subscriptions. The Spanish guide says a free edition covers real-time protection, on-demand scans and AI-based detection, with only advanced features such as Controlled Folder Access and full Endpoint Manager integration requiring E5 or Microsoft Defender for Cloud. Anyone budgeting a rollout should verify current terms with Microsoft before relying on either claim.
Why it matters
Linux servers have long leaned on signature-only scanners such as ClamAV, and both guides argue that AI-assisted malware changes that calculus. If mdatp delivers full-scan CPU overhead in the low single digits and EDR telemetry that lands in the same console as Windows and macOS endpoints, it becomes an attractive option for mixed estates already running Microsoft security tooling. The caveats are just as clear: the two posts disagree on licensing, idle memory use and which competitors were beaten, and neither offers a reproducible methodology. Treat them as deployment recipes and starting points for your own evaluation rather than settled performance data.
- #security
- #linux
- #microsoft-defender
- #endpoint-security
- #devops