Why migrate from PRTG to Zabbix? #

We have been using PRTG Network Monitor to monitor UPS (uninterruptible power supply) units in a portion of our facility via SNMP.
It was stable and we had no particular complaints, but as we approached the license limit on sensor count
and undertook a cost review, we decided to consider a migration to Zabbix.

Zabbix is open source with unlimited host and metric counts. It also has rich official plugins with Grafana.
There was no reason not to migrate. The issue was the migration work itself.

This time we tried Claude Code.
We passed a PRTG backup file as-is and handed off everything from parsing to script generation and execution.
As a result, migration of all 12 devices was completed in approximately 30 minutes. However, there was one stumbling point, so we’re recording that as well.

Migration scope this time #

Analyzing the PRTG backup (PRTG Configuration.dat), we found this configuration.

Migration architecture #

Normal migration requires a PRTG API server to be running, but this time we took a method to access the backup file directly.
The .dat file is just XML under the hood, so it can be parsed offline without a server.

PRTG Configuration.dat  （XML file）
        │
        ▼  Pass to Claude Code only
  ┌─────────────────────────┐
  │      Claude Code        │
  │  ① Parse XML            │
  │  ② Extract device list  │
  │  ③ Generate Python script│
  │  ④ Execute Zabbix API   │
  └─────────────────────────┘
        │
        ▼
  Zabbix（Host, Interface, Template registration complete）

Actual work log #

Instructions given to Claude Code #

We launched claude on the terminal and passed the following single instruction.

Please analyze the PRTG backup file and migrate it to Zabbix.

File: ./PRTG\ Configuration.dat

Steps:
1. Parse XML and output all device list (name, IP, sensor type) to CSV
2. Review device list, provide migration plan summary
3. Use Zabbix API (http://zabbix.example.local) to execute the following:
   - Create host group "Migrated-UPS"
   - Register each device with SNMP interface (port 161)
   - Apply templates
4. Verify import results and output difference report

Zabbix API token: xxxxxxxxxxxxxxxx

Step ① XML parsing and device extraction (approximately 5 minutes) #

Claude Code immediately read the PRTG Configuration.dat file and automatically generated and executed a Python script to parse the XML.
It took less than 5 minutes to output device names, IP addresses, and sensor types to CSV.

import xml.etree.ElementTree as ET
import csv, re

with open("PRTG Configuration.dat", encoding="utf-8-sig") as f:
    content = f.read()

# Extract device name, IP, sensor type via regex
devices = re.findall(
    r'<name[^>]*>\s*([A-Za-z0-9_\-\.]+(?:-ups-\d+|UPS-\d+)[^<]*)\s*</n',
    content
)

# CSV output
with open("prtg_export.csv", "w", newline="") as f:
    w = csv.writer(f)
    w.writerow(["name", "ip", "interface_type"])
    for name, ip in zip(device_names, device_ips):
        w.writerow([name, ip, "SNMP"])

Step ② Zabbix import script generation and execution (approximately 10 minutes) #

Once the CSV was ready, Claude Code continued to generate and immediately execute an import script that called the Zabbix API.
There was zero overhead of manually copying and pasting code and debugging it ourselves. This is Claude Code’s greatest strength.

import requests, csv, json

ZABBIX_URL  = "http://zabbix.example.local/api_jsonrpc.php"
API_TOKEN   = "xxxxxxxxxxxxxxxx"
TEMPLATE    = "APC UPS by SNMP"
HOST_GROUP  = "Migrated-UPS"

def zabbix_call(method, params):
    payload = {"jsonrpc": "2.0", "method": method,
               "params": params, "id": 1,
               "auth": API_TOKEN}
    r = requests.post(ZABBIX_URL, json=payload)
    return r.json().get("result")

# Create host group
group_id = zabbix_call("hostgroup.create", {"name": HOST_GROUP})["groupids"][0]

# Get template ID
tmpl = zabbix_call("template.get", {"filter": {"host": [TEMPLATE]}})
tmpl_id = tmpl[0]["templateid"]

# Read CSV and register hosts
with open("prtg_export.csv") as f:
    for row in csv.DictReader(f):
        zabbix_call("host.create", {
            "host": row["name"],
            "interfaces": [{"type": 2,  # SNMP
                            "main": 1, "useip": 1,
                            "ip": row["ip"], "dns": "",
                            "port": "161"}],
            "groups":    [{"groupid": group_id}],
            "templates": [{"templateid": tmpl_id}],
        })

Stumbling block: Voltage anomaly alerts triggered in mass #

Right after import, the Zabbix alert screen turned bright red.
“Input voltage too high” triggered simultaneously on all 12 devices.

Issue modification instructions and re-registration #

We issued additional instructions to Claude Code.

Voltage anomaly alerts are being triggered.
The applied template appears to be in 200V system settings.

Please perform the following:
1. Unlink the template from all 12 hosts
2. Search for the correct template "APC UPS by SNMP (100V)" and apply it
   ※ If it doesn't exist, duplicate the existing template,
     modify the input voltage trigger thresholds to 90-110V, and apply it
3. Confirm the alerts have been resolved and report the results

Claude Code automatically performed template unlink → threshold correction → re-application.
This correction work, including the initial import, was resolved within 15 minutes.

Summary of results #

What could be automated and what could not #

Benefits of using Claude Code #

1. Code is generated and executed on the spot.
   Traditional AI chat required a loop of “generate code → paste and run → paste error and ask again”.
   Claude Code executes directly on the terminal, reads errors itself, and corrects them.
2. Verifies API documentation on its own.
   When we asked “What are the parameters to register an SNMP interface via Zabbix API?”, it researched this itself.
   There was almost zero overhead in reading specifications.
3. Honestly communicates what it cannot do.
   When we asked for complete 1:1 migration of PRTG-specific sensor settings,
   it appropriately drew the line: “Due to conceptual differences with Zabbix, 1:1 migration is difficult. Manual work recommended for this part.”
4. Can recover from failures with a one-line correction instruction.
   The template issue was resolved with just “fix this”.
   There was virtually no debugging stress.

Next preview: Network equipment edition #

This time was smooth because all devices were “identical configuration” APC UPS.
Next time, we plan to work with a 2023 backup that includes switches, routers, wireless APs, and virtual machines,
testing multi-template distribution and SNMP community string handling as well.

• 10 switches
• 4 routers
• 4 wireless APs
• 4 HyperV hosts + 10 virtual machines