The test bench
Server tests run on real, paid infrastructure, not on vendor-sponsored credits. Right now the bench is one DigitalOcean droplet in Frankfurt on the $6 plan (1 vCPU, 1 GB RAM, 25 GB SSD, Ubuntu 24.04). It is deliberately small: if a tool works there, it works on anything a small team would buy, and if it does not, we can show exactly where it runs out.
When a comparison needs a second server or another provider, we buy it for the test and say so in the article.
MCP servers are the exception to the server rule: they are tested where people actually run them, on a developer machine (currently macOS on an Apple M1 Pro). We talk to each server directly over the protocol with our own client script, record the tool list it returns, time every call and measure memory. The article names the machine and the exact version tested.
What we measure and how
| What | How we measure it |
|---|---|
| Install and deploy time | date before and after each step, on the server itself |
| Memory | free -m for the whole machine, docker stats per container, sampled during builds |
| Disk | df -h before and after installs and builds |
| Cost | The plan price on the vendor's pricing page on the day, then the real invoice once the month closes |
| Whether it works | An HTTP request from outside the server, not the green badge in the dashboard |
Two kinds of articles
Reviews are full hands-on tests. Everything in them was done by us on the bench: the commands, the screenshots, the timings.
Blog posts are shorter and answer one question. Each has at least one fact we checked ourselves that day, shown in a "Checked" box with the date and the source: a command we ran, a pricing page, a line in the official docs. When a blog post uses numbers from one of our reviews, it links to the review they came from.
Dates and updates
Prices, docs and interfaces change. Every number in an article carries the day it was measured or checked. A review is published with first-day numbers and updated after 7 and 30 days of running; the article shows an "updated" date when that happens. If a vendor changes pricing or behaviour and you notice before we do, tell us.
Mistakes stay in
If we typed the wrong directory, forgot a firewall rule, or ran out of memory, the article says so, because you will probably hit the same thing. A test that only shows the happy path is an advert.
Writing tools
We use AI tools to help draft and edit. The measurements, commands, screenshots and verdicts come from our own tests and from sources we checked on the date shown.
Money
We pay for the servers and plans we test. Some links are affiliate links, always marked, and the verdict is written before we look at what a program pays. Tools that pay us nothing are covered the same way. Details: how we make money.
Corrections
Found an error or a number that no longer holds? Write to hello@fewservers.com. We fix it and note the change in the article.