Endpoint protection proposals keep repeating the same three acronyms, and vendor materials often blur them together. EDR and XDR are technology: the software collects the data and carries out the response. MDR is a service: analysts review the alerts and act as agreed.
The difference starts to matter when you buy. If you buy a product, you get a tool, and your own people have to deal with the alerts. If you buy a service, you have to decide in advance what the provider may do without your approval. Below we go through the three terms first, then look at how the work and the decisions are split.
EDR: detection and response on the endpoint
EDR (Endpoint Detection and Response) is built on an endpoint agent and a central analysis system; the agent runs on workstations and servers. It continuously records what happens on the machine: which process started and what started it, with which command-line parameters, which file it wrote, which setting it changed, and where it opened a network connection. This telemetry goes to the management console, where rules and behavioral models analyze it.
Detection works on chains of events. An office document launches a command interpreter, the interpreter runs an encoded script, and the script connects to an unknown address. Each step is ordinary on its own, but one after the other they are suspicious. The article on choosing an EDR compares it with classic antivirus in detail.
What it can do on the endpoint
- Stops a process and quarantines the file that belongs to it.
- Isolates the machine from the network. The machine then keeps a connection only to the management console, so it can still be investigated, but the attacker cannot move on from it.
- Opens a remote investigation session in which the analyst can collect files and run commands.
- Searches the stored telemetry retroactively, for example for a newly published indicator of compromise (IoC).
EDR sees what happens on the machine its agent runs on. If an attacker signs in to a cloud service with a stolen password, or moves through a device that cannot take an agent (a network device, a printer, a legacy system), none of it shows up in the console.
XDR: several sources in one analysis
XDR (Extended Detection and Response) analyzes EDR telemetry together with other sources. Typical sources are the email filter, the identity provider's sign-in logs, the firewall, and cloud workloads. The goal is to bring signals that appear in different places together as a single event on a shared timeline.
A simplified example: a phishing email arrives, a suspicious process starts on the recipient's machine after the attachment is opened, and an hour later someone signs in to a cloud application with the same account from an unusual location. With separate tools, this would be three alerts in three consoles, and it would be up to the analyst to connect them. XDR links the three into one incident, which shows that the account and the machine belong to the same attack and that both need action.
There are two approaches on the market. One connects data from the vendor's own products, and is often called native XDR. The other also accepts sources from third-party vendors through standard interfaces; the name open XDR has caught on for this. In practice the line between them is blurry, so ask specifically which sources can be connected and how much detail the data from each one carries.
Where the SIEM fits
A SIEM is a general log collection and correlation system: it accepts almost any source, keeps data for a long time, and is usually where compliance reports are produced. XDR works with fewer sources, but comes with prebuilt detections and built-in response actions. Many organizations have both, and the XDR incidents are forwarded to the SIEM as well. The steps of a SIEM deployment are covered in a separate article.
MDR: when a provider monitors the alerts
EDR and XDR produce alerts, and someone has to review them day and night. A smaller IT team can rarely take this on, because it needs an on-call rotation, experienced analysts, and in-depth knowledge of the platform. MDR (Managed Detection and Response) provides this work as a service.
MDR is built on an EDR or XDR platform. It can be delivered by the platform vendor itself, or by an independent provider that works on the platforms of one or more vendors. What the service includes depends on the description in the contract; it is usually made up of the following:
- continuous monitoring and triage of alerts, meaning filtering out the false positives;
- investigation of cases that look real: what happened, which machine and account are affected, how far the attacker got;
- threat hunting, where analysts search the telemetry for known attack patterns without waiting for an alert, for example lateral movement;
- carrying out the agreed response actions, or a notification when an action needs the organization's approval;
- a written report on each case and a regular summary.
In proposals, the word “monitoring” can mean many different things. Sometimes it only means that the provider forwards severe alerts by email, while the investigation and the response are left to the organization.
Who decides on isolation at 2 a.m.
At 2 a.m., the analyst sees a process on a file server rewriting the files in the shared folders one after another, which points to encryption. Isolating the server could stop the damage, but the server stays unreachable until someone lifts the isolation. If the analyst first has to get approval by phone and the on-call person on the list does not pick up, the encryption carries on in the meantime. If, on the other hand, the provider can isolate anything on its own, a false positive could leave an important system down in the morning.
The answer is authorization set down in advance, by device group and by action. One possible split:
| Device group | Stopping a process, quarantine | Network isolation |
|---|---|---|
| Workstations | The provider on its own, with notification afterward | The provider on its own, with immediate notification |
| General servers | The provider on its own, with immediate notification | On its own for a confirmed, active attack, otherwise with approval |
| Critical systems (domain controller, ERP system, servers that support production) | With approval | With approval, plus a recorded rule for what happens if no one can be reached within the set time |
The table needs a contact chain to go with it, with names, phone numbers, and deputies, and someone has to know how to lift an isolation. If the XDR also sees an identity source, disabling an account or revoking its sessions gets a separate row, because disabling an executive or service account has different consequences from isolating a workstation. Test the authorization in an exercise, before a real night reveals that one of the numbers on the list no longer works.
What stays with the organization
An MDR contract hands monitoring and part of the response over to the provider. In the four areas below, the provider cannot do its job well without the organization's own work, and the responsibility stays with the organization too.
Coverage
The analyst sees only the machines that run an agent. New machines have to be added when they are set up and decommissioned ones removed, and someone has to keep a list of the machines that cannot take an agent. Protecting those, for example by separating them on the network, is up to the organization.
Business context
The provider has no way of knowing by itself which server is critical, which remote administration tool the IT team uses legitimately, or when planned maintenance is scheduled. Without this information it either raises needless alerts or asks for exceptions that an attacker could hide behind as well. The organization approves the exceptions and reviews them from time to time.
The incident beyond the endpoint
Isolation is only the start of an incident. Restoring from backup, changing passwords, internal and external communication, and management decisions all fall to the organization, and so does notifying the authorities. For organizations within the scope of the NIS2 Directive, reporting significant incidents remains the organization's obligation even when monitoring has been outsourced. The article on the first 24 hours after an incident describes what to do in the first hours.
Acting on the recommendations
MDR reports often point to an unpatched vulnerability, a permission that is too broad, or logging that has been switched off. Fixing these is a job for IT operations. Put the recommendations into your own task tracker with an owner and a deadline, and check at the next report which ones have been done.
What to clarify in an MDR contract
Proposals use similar names for the service but differ a great deal in the details. Settle these questions in writing:
- What response time means: the time from the alert to the analyst's first action, to the notification, or to the response action, and how long it is at each severity level.
- Authorization for response actions by device group, as in the previous section, and how it can be changed later.
- How and in what order notification happens: phone, email, or ticket, who gets called first, and what happens if no one can be reached.
- Which sources are included: only the endpoints, or identity, email, and cloud services as well.
- Whether threat hunting is part of the service, how often it runs, and whether it comes with a report.
- Where the investigation ends. In many contracts, detailed forensic analysis and recovery are already a separate engagement, and it is better to know this up front.
- The data: where the telemetry is stored, for how long, who can access it, and what happens to it when the contract ends. Because telemetry also contains personal data (usernames, opened files, browsing traces), a data processing agreement is needed as well.
- How tuning is handled: who can request an exception and who can approve one, and how the reason for an exception can be traced back.
Related service: Endpoint protection (EDR/XDR) →