CrowdStrike applications (WIP)
Reads what CrowdStrike Falcon sees on your servers and turns the part of it that matters to a business application register into Applications: business software such as BI, ERP/CRM, collaboration and email platforms, your own containerised software, and the data platforms behind it, such as databases, message brokers and web and application servers. Each product becomes one application, linked to the endpoints it runs on.
| Type name | crowdstrike_applications |
| Creates | Applications |
| Runs on | A sensor with outbound reach to the Falcon API |
| Fields written | names, type, hw_vendor, hw_product, hw_version, application_hosts, and the csapp_* fields |
Where the evidence comes from
Four sources, each switched on or off in the scan template. A source the API client cannot read is skipped with a warning in the scan log; the others still run. The scan fails only when no enabled source can be read, or when every source is switched off.
| Source | What it shows | Shown as |
|---|---|---|
| Installed software (Falcon Discover) | The software installed on each host, as Windows or the package manager reports it | Installed |
| Running containers (NG-SIEM) | The image every container ran from, per host, with its Kubernetes namespace. Software shipped in an image is installed nowhere on the host, so the installed software never shows it | Container |
| Running server processes (NG-SIEM) | Known server processes starting: database engines, web servers, Exchange, SharePoint. Finds what was installed outside a package manager, such as Oracle Database, and confirms what actually runs | Running |
| Listening ports (NG-SIEM) | The ports each application listens on. Ports only add to an application found otherwise; they never create one | Ports |
The three NG-SIEM sources search back over the NG-SIEM look-back window, 7 days by default. A running container re-sends its details, so it is found even when it started long ago. A process or a port is only recorded when it starts, so a database that has run untouched for months may not show as running: the installed software still finds it.
What counts as an application
A server's software inventory is mostly not applications: security agents, drivers, runtimes, browsers, management consoles and operating system packages. Each installed name is judged in this order:
- Names containing one of the template's Never include entries are skipped.
- Names containing one of the Always include entries become that application, named as you wrote it. Use this for line-of-business products the platform does not know, such as a core banking system.
- Tooling around a product is skipped: drivers, clients, agents, SDKs, runtimes, management studios, setup files, updates and OS libraries.
- Known products are recognised by name and folded together. Qlik Sense, for example, arrives as its proxy, repository, scheduler and licence services, and becomes one Qlik Sense application.
- Anything else CrowdStrike files under a business category (database, web server, business intelligence, collaboration, ERP, CRM and similar) is kept under its own name, without its version number.
- Everything else is skipped.
IT infrastructure (backup, monitoring, container platforms, security tooling and software delivery such as CI servers and artefact repositories) runs the estate rather than the business, so it is skipped unless Include IT infrastructure is on.
The platform knows around 150 products by name, image or process: databases, web and application servers, message brokers, data and AI platforms, business, collaboration and content management software. Where it has the product's logo, the Applications list and the application's page show it beside the name.
Containers
A container image is judged by its repository, without its registry and tag:
- Never include and Always include apply as for installed names. Always
include gathers every image whose repository contains the name into one
application, so
billingmakesbilling-api,billing-webandbilling-workerone billing application. - Known products are recognised:
postgres,bitnami/postgresqland an image of PostgreSQL from any registry are all PostgreSQL, the same application as PostgreSQL installed on a server. - Cluster plumbing is skipped: Kubernetes components, networking, storage drivers, ingress and certificate controllers, node agents, log shippers, init containers and debugging shells.
- Any other image becomes an application of its own, named by its repository, of type Containerised application. In a container estate this is how your own software shows up. Turn off Own container images to keep only known products.
Each image repository is its own application, so a system built from many services, or pushed to a separate repository per environment, becomes many applications. Use Always include with the system's name to gather them into one.
Processes and ports
Only a product the platform knows by its process becomes an application from a running
process; a process such as java or node says nothing about what runs. An engine that
a desktop tool starts for itself, such as Power BI Desktop's, is not a server. An older
MariaDB still runs as mysqld, so on a host with MariaDB that process counts as MariaDB.
A port is added to an application when its container opened it, or when it is one of the product's own ports (1433 for SQL Server, 5432 for PostgreSQL) on a host the product runs on. To search for a process the platform does not know, add its name to Also search these process names and to Always include.
A database counts only where its server components are installed. SQL Server
Management Studio, a client driver or an operating system prerequisite package
(oracle-database-preinstall-19c, for instance) does not make a host a database server.
A web application deployed onto IIS or Tomcat appears here only as the platform it runs on. Add such applications with Always include when they have an installed component or a process of their own, or register them by hand and link their hosts. Software in containers is found, see Containers.
One application per product
| Field | Holds |
|---|---|
names, hw_product | The product, e.g. Microsoft SQL Server |
hw_vendor | The vendor |
type | What kind of application it is: Business application, Business intelligence, Collaboration, Email, Database, Message broker, Data platform, AI platform, Web server, Application server or Containerised application. With Include IT infrastructure on, also Backup, Monitoring, Container platform, Security or Software delivery |
hw_version | Every version installed across its hosts |
application_hosts | The endpoints it runs on, most critical first, then by name, up to 100 |
csapp_product, csapp_vendor, csapp_kind | The product, vendor and kind as this integration last saw them, kept even when the shared fields were corrected by hand |
csapp_product_key | The stable key the product is matched by on later runs |
csapp_categories | The CrowdStrike software categories of its installed names |
csapp_versions, csapp_platforms | The versions and host platforms seen, comma-separated |
csapp_host_count | How many hosts it runs on, without a limit |
csapp_hosts | Each host with its versions, its evidence and its ports, e.g. DB01 (16.0.4262.2) - Installed, Running - ports 1433. Lists up to 500 hosts, then says how many more there are |
csapp_components | The installed names, images and processes folded into it |
csapp_evidence | What shows it on its hosts: Installed, Container, Running |
csapp_ports | The ports it listens on across its hosts |
csapp_namespaces | The Kubernetes namespaces its containers ran in |
csapp_highest_host_criticality | The highest CrowdStrike criticality among its hosts |
csapp_last_seen, csapp_last_used | The latest sighting on any host, and the last use CrowdStrike recorded |
type, hw_vendor and hw_product are written only when empty, so a correction made
in the register stays. The application's own criticality is never written: it is a
business judgement, and the host criticality is shown separately as a hint.
Each endpoint gets a Hosted applications list, the applications that name it among their hosts.
A product keeps its application across runs through its product key. An application someone created by hand under the same name is adopted rather than duplicated.
Requirements
The integration is outbound-only: it is a scheduled API client, so nothing is installed on the CrowdStrike side and nothing runs on the endpoints themselves.
Network access
| From | To | Protocol | Port |
|---|---|---|---|
| The sensor running the integration | The Falcon API host for your region, e.g. api.us-2.crowdstrike.com | HTTPS over TCP | 443 |
Access in CrowdStrike
| Need | Why |
|---|---|
| An API client with the Falcon Discover: Read scope (shown as Assets: Read in Exposure Management) | The installed software is read from Falcon Discover |
| The same client with the Investigate (LogScale) read scope | Containers, processes and ports are searched in NG-SIEM |
| Falcon sensors with container visibility on the hosts that run containers | The sensor records the containers. Hosts without it show no containers |
A scope added to an existing client takes effect only on a token minted after the change. A client with only one of the two scopes works: the sources it cannot read are skipped, and the connection test names them.
In this deployment
| Need | Why |
|---|---|
| A registered, healthy sensor | The integration runs on a sensor, on the platform's schedule. See Sensors |
| Endpoints already discovered by CrowdStrike | Software is kept only on hosts that are already endpoints of the tenant, matched by their Falcon sensor ID. Schedule this after the CrowdStrike scan |
When one Falcon account holds several organisations' hosts, this integration keeps only the hosts that are endpoints of the tenant. That separates the organisations only if the CrowdStrike integration of each tenant imports its own hosts and no others.
Configuration
| Field | Required | Notes |
|---|---|---|
| Name | Yes | Label for this instance |
| API URL | Yes | Region-specific Falcon API base, e.g. https://api.us-2.crowdstrike.com |
| Client ID | Yes | Falcon API client ID |
| Client Secret | Yes | Falcon API client secret. Write-only: once saved it shows as a locked box with Replace, and is sent only to the sensor |
| Proxy settings | No | http://host:port |
| Sensor | Yes | Sensor that will run the integration |
| Ignore SSL | No | Skips certificate validation. Off by default |
Test connection checks each enabled source separately: it asks Falcon Discover for one application record on the hosts in scope, and submits a short NG-SIEM query job. It passes when at least one enabled source is readable and lists the ones that are not.
Its scan template adds:
| Option | Default | Effect |
|---|---|---|
| Hosts | Servers and domain controllers | Which hosts are read, by every source: servers and domain controllers, servers only, or all hosts |
| Installed software | On | Reads the installed software (Falcon Discover) |
| Running containers | On | Reads the images containers ran from (NG-SIEM) |
| Running server processes | On | Reads which known server processes started (NG-SIEM) |
| Listening ports | On | Adds the ports each application listens on (NG-SIEM) |
| NG-SIEM look-back window | 7 days | How far back NG-SIEM is searched |
| Also search these process names | Empty | Comma-separated process names to search for besides the known server processes |
| Include IT infrastructure | Off | Also keeps backup, monitoring, container platforms, security tooling and software delivery |
| Also count these CrowdStrike categories | Empty | Further CrowdStrike software categories to treat as business software |
| Always include | Empty | Comma-separated product names to keep whatever CrowdStrike files them under; matches installed names, image repositories and process names |
| Never include | Empty | Comma-separated names to skip, in installed names, image repositories and process names |
| Own container images | On | Makes every container image not otherwise recognised an application of its own |
| Page size | 1000 | Installed-software records per request, 1 to 1000 |
See also
- CrowdStrike: the endpoint inventory this depends on
- Assets: how applications sit alongside endpoints and users
- Integrations overview: the shared configuration model