A listing on Few Servers has two layers. Most of what you see is listed: facts we read from the project's or the company's own pages. A smaller part is tested: things we installed, called or rented ourselves. The page always says which is which.
Listed: where the facts come from
MCP servers come from the official MCP Registry, which we crawl and count ourselves; the home page shows the count with the crawl date. Install details on a server page are the ones in its registry record, and star counts are read from the project's own GitHub repository, each with the day it was read. A server gets its own page when it is tested by us, published under a vendor's own namespace, fully documented in the registry, or popular on GitHub; the others stay in the lists with a "Claim" link.
Server providers are read from the company's own website, pricing page or public API. Every fact (product types, data-centre countries, lowest listed price, payment methods, support, status page, year founded) is stored with its source address, the date we read it and the words it was taken from, and the site will not build if a fact has no source. Prices are the lowest price the company listed on that date, in the currency it listed; they are not a quote and may leave out tax.
Three things we do not do: copy facts from other directories, fill a field we could not read (it stays empty), or rate a listing ourselves.
Submissions, claims and corrections
Anyone can submit an MCP server or a provider for free, claim an existing listing or suggest a correction. A person reads every submission and checks it against the project's or company's own pages before anything changes. The contact email you give is never shown.
Tested: what the mark means
"Tested by Few Servers" appears only where we ran the thing ourselves, and it links to the write-up.
- MCP servers are tested where people run them, on a developer machine (currently macOS on an Apple M1 Pro). We talk to the server over the protocol with our own client script, record the tool list it returns, time the calls and measure memory. The write-up names the machine, the version and the date.
- Servers and self-hosted tools are tested on infrastructure we pay for, not on vendor credits. The usual bench is one DigitalOcean droplet on the $6 plan (1 vCPU, 1 GB RAM, 25 GB SSD, Ubuntu 24.04). It is deliberately small: if a tool runs there, it runs on anything a small team would buy. When a test needs another server or provider, we buy it and say so.
| 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 |
| Disk | df -h before and after installs and builds |
| Cost | The plan price on the vendor's pricing page on the day, then the invoice |
| Whether it works | An HTTP request from outside the server, not the green badge in the dashboard |
A tested mark is dated. It says what happened on that version on that day, not that the product is good, and it cannot be bought.
Dates and updates
Prices, docs and registry records change. Every fact shows the day it was read, and every test the day it was run. If something on a listing no longer holds, use "Suggest a correction" on the page or write to us.
Guides and reviews
The hands-on reviews and blog posts published before the site became a directory follow the same rules: commands we ran, numbers we measured, mistakes left in, and a "Checked" box with the source and date for anything we read rather than ran.
Writing tools
We use AI tools to help collect, draft and edit. Facts come from the sources shown next to them, and measurements from our own tests.
Money
Listings are free and no vendor pays us for a listing or a link today. A paid placement, if we sell one, will be labelled Sponsored. Details: how we make money.
Corrections
Write to hello@fewservers.com. We fix it and the listing shows the new read date.