Muster.

guides

Scanning

Muster's agents attack across several engines, on demand or on a schedule. This page covers what runs, how to launch and schedule scans, and the one safety setting to understand before you point it at production.

what you'll learn

  • The engines and what each one produces
  • How to launch a scan — one asset, all assets, or the AI probe
  • Passive vs. active web scanning, and why it matters
  • How to schedule continuous scans and watch them run
  • How to control what the GitHub App posts on your pull requests

The engines

Each asset runs the engine(s) that fit it. Findings are tagged with the engine that produced them:

parametertypedescription
code scanningcodeSource-code analysis on connected repositories.
web appwebDynamic testing of web apps, including the adaptive AI probe.
depsdepsDependency / known-vulnerability checks.
cloudcloudCloud posture across AWS, Azure, and GCP.
DNSdnsHourly DNS drift monitoring per web app; each app can limit it to hijack signals or turn it off.

Launching a scan

Scan controls live on the org's assets tab. You can run everything at once or target a single asset:

  1. 1

    Scan everything

    Scan all assets launches every connected asset in the org. Good after onboarding, or a broad re-check.

  2. 2

    Scan one asset

    Use the row's Scan button. Repos also have Deps (dependency scan), and web apps have ✦ Probe (below).

  3. 3

    Or from code / CI

    Call POST /api/v1/orgs/:id/scan from a pipeline step so every deploy is tested, or trigger it from your editor with the MCP server. See the API reference.

The adaptive probing agent

On a web-app asset, ✦ Probe turns loose an AI agent that adaptively probes the target — reasoning about the app to hunt logic flaws, IDOR, and auth bypasses a fixed scanner would miss. It uses any authentication and context docs you configured to reach deeper into the app.

Passive vs. active web scanning

Web-app scanning has two modes, and the difference matters when the target is production:

  • Passive (safe) — observes and analyses traffic without sending attacks. Safe to run anywhere.
  • Active (intrusive) — sends real attack traffic to prove exploitability. More thorough, but it exercises the app for real.
⚠
Active scanning is gated: each web app has an authorize checkbox and requires the ownership attestation. Point it at staging first if you're unsure, and prefer a test account for authenticated runs.

Scheduling continuous scans

The real value is continuous coverage. Each organization has one scan schedule, under Settings → Scanning (/orgs/<your-org>/settings?tab=scanning). Only org owners and admins can change it; other members see it read-only.

  1. 1

    Turn on scheduled scans

    Flip the Scheduled scans switch on. Nothing changes until you click Save, and Cancel puts back the saved settings.

  2. 2

    Pick a frequency

    Choose Every hour, Every 6 hours, Daily, or Weekly. Weekly scans need the Indie plan or above; anything more frequent than weekly needs the Business plan or above. Frequencies your plan doesn't include show a lock and link to Billing so you can upgrade.

  3. 3

    Choose what to scan

    Under What to scan, pick All assets, Web apps only, Repos only, or Contracts only. The preview underneath shows how many assets each run covers (for example, “Will scan 11 assets: 8 web apps, 3 repos”). If the scope matches nothing, add assets on the Assets page first.

  4. 4

    Set the web app scan intensity

    Passive (safe) crawls and runs non-intrusive checks. Active (sends attack traffic) runs attack payloads on every scheduled run, and the AI agent then probes the app for logic and access-control flaws. Choosing active asks you to confirm first. Intensity only affects web apps.

  5. 5

    Save

    Click Save. The panel shows the next and last run in your local time, with relative times such as “in 5h” or “2d ago”. A newly enabled schedule runs within a minute, then repeats at the chosen frequency. Run now scans the saved scope once, with the saved intensity, without changing the schedule.

GitHub pull requests

With the Muster GitHub App connected, every pull request on a repo with scan PRs checked gets a security scan of the files it changes. Results always show in Muster. What the App posts on GitHub is up to you, one action at a time, under Settings → Integrations → GitHub:

  • Status checks: a Muster / PR Scan check on each commit with inline findings. Blocking gate mode needs this on.
  • PR summary comment: one comment per pull request, updated in place after each scan.
  • /muster commands: replies and reactions to /muster commands, and fix PRs opened by /muster fix. When off, commands are ignored.
  • GitHub issues: opening issues from findings, and closing them when a finding is resolved.

Each action can be off, on, or on a date: it starts at midnight (your time) on the day you pick and posts nothing before then, so you can tell your team ahead of time. The block at the top sets the account default; open posts on GitHub under any repo to override it for that repo, or set it back to account default. A good rollout is scans only first, then status checks, then comments and commands.

ℹ
New connections start quiet: PR scans run, but nothing is posted to GitHub until you turn an action on. Connections made before these controls existed kept everything on, so nothing changed for them; turn off anything you don't want. A check Muster has already started is still completed if you turn status checks off, so a pull request is never left with a check that never finishes.

Watching scans

Follow runs under the org's scans tab (/orgs/<your-org>/scans): a live scan activity feed, a scan history you can click into, and per-run detail at /orgs/<your-org>/scans/<id>. Prefer to poll programmatically? Use GET /api/v1/orgs/:id/scans.

Re-testing after a fix

There's no per-finding “re-test” button — you re-scan the asset. After shipping a fix, run Scan (or POST /api/v1/orgs/:id/scan with { "all": true }). A finding the agents can no longer reproduce stops recurring; if it ever returns, it comes back flagged regressed. Full detail in Findings → Confirming a fix.