Best Docker security scanners (2026)

Trivy (Apache-2.0, 36,000+ GitHub stars) is the de-facto standard open-source container security scanner. A single binary covers CVE scanning across OS packages and 20+ language ecosystems, IaC misconfiguration detection (Terraform, CloudFormation, Kubernetes, Helm, Dockerfile, Ansible), secrets scanning, license auditing, and SBOM generation (CycloneDX, SPDX). It runs fully locally — CVE DB downloaded and cached — with no account and no cloud calls.

Where Trivy is limited: it scans images, filesystems, and manifests. It does not scan the live running box outside-in, does not check what ports are actually reachable from the internet due to Docker's iptables rules, and has no awareness of AI agents or MCP servers. It belongs in your build pipeline as the image-layer gate; it does not replace a deployment-context scan.

Note on the March 2026 supply-chain incident: Threat actors compromised Aqua Security's CI/CD pipeline and force-pushed 76 of 77 trivy-action release tags to credential-stealing malware on GitHub. (In a separate prong of the same attack, two malicious image tags were pushed to Docker Hub.) Pin Actions to immutable commit SHAs in production rather than mutable version tags.

Grype (Apache-2.0, ~12,370 GitHub stars) by Anchore is a fast, single-binary CVE scanner for container images and filesystems. Like Trivy it covers OS packages and 20+ language ecosystems, supports EPSS and KEV scoring for risk prioritization, and handles OpenVEX for filtering non-exploitable findings. No account required; runs fully locally.

Grype's differentiator is its tight integration with Syft: teams that generate SBOMs at build time and scan them separately (decoupling SBOM generation from vulnerability lookup) tend to prefer Grype. For teams who want IaC + image scanning in a single tool, Trivy has the edge. For CVE-only pipelines, Grype is equally capable and lighter.

Docker Scout ships inside Docker Desktop and the Docker CLI, making it the lowest-friction starting point for teams already on Docker Hub. It generates SBOMs, matches against 23+ advisory sources, and surfaces remediation suggestions (specific base image recommendations, not just CVE IDs). The policy evaluation engine can enforce security standards across a team.

The meaningful trade-off: Scout is cloud-backed. Image analysis data is sent to Docker's infrastructure, a Docker account is required for Hub-linked features, and the free Personal plan is limited to 1 Docker Scout-enabled repository. For teams with strict data-residency requirements or self-hosted environments where nothing should leave the box, Trivy or Grype are the standard alternatives. Scout also does not scan the host OS, IaC files, or running workloads — scope is image-only.

Dockle (Apache-2.0, ~3,260 stars) fills a specific gap that Trivy and Grype do not: it checks built container images against the 11 CIS Docker Image Benchmark checkpoints (CIS-DI-0001 through CIS-DI-0011) plus its own proprietary best-practice checks. This covers things Trivy's Dockerfile linting misses — for example, whether a setuid/setgid binary ended up in the final image, whether sensitive files were accidentally COPY'd in, or whether an image runs as root when it should not.

Dockle is a linter, not a CVE scanner — it does not match packages against vulnerability databases. Run it alongside Trivy or Grype: Trivy/Grype for CVEs, Dockle for CIS-DI compliance at the image layer. Local-first, no account, single binary.

Lynis (GPL, by CISOfy) is a host-based security auditor — not a container scanner. It runs directly on the Linux host and checks hundreds of configuration points: SSH hardening, PAM, boot loader, firewall settings, file permissions, package freshness, and more. It produces a hardening index (a number) and a prose report with per-finding remediation guidance. Compliance modes for PCI DSS, HIPAA, and ISO 27001 are available.

What Lynis does not do: it does not scan container images for CVEs, does not check what Docker has opened through iptables, does not assess AI agents or MCP configs, and does not produce a shareable audit record mapped to control frameworks. Its report is a human-readable text file, not a structured evidence artefact you can feed into a compliance workflow. For the host layer, Lynis is thorough; for the whole box, it is half the picture.

Keelix is the only scanner in this roundup that operates at the deployment-context layer. It runs on the live box and covers what the others do not: host OS configuration checks (mapped to CIS, SOC 2, and ISO 27001 controls), outside-in exposure (exactly which ports are reachable from the internet, including what Docker opened through iptables), and the AI agents and MCP servers running on the machine — auto-approval risk, plaintext secrets in MCP config, unvetted server publishers, and the lethal trifecta.

All of that folds into a single 0–100 posture score you can gate CI on: set a threshold, and the deploy blocks if anything pushes the score below it. Findings map to control frameworks and are exported as shareable audit records — JSON, HTML, or scan record format.

Where Keelix is honest about its limits: it does not perform deep CVE scanning of image layers (Trivy and Grype do that better), it does not generate SBOMs, and it does not lint Dockerfiles for best-practice issues (Dockle and Trivy do that better). The intended workflow is run both in sequence — Trivy in the image build stage, Keelix in the deploy-to-staging stage.

Apache-2.0 CLI, free, nothing leaves the box.