Published on : 2026-09-01 12:08
I recently came across gopher-mcp, a project that lets AI
assistants browse Gopher and Gemini:
(HTM) We're dommed - BBS [gemini]
At first glance, this sounds pretty cool.
An MCP server for the small web!
It supports Gopher menus, text files as well as Gemini resources.
It has timeouts, response limits, caching, host allowlists and
other sensible security features. It even returns structured data
specifically designed for LLMs.
Technically, it's a neat project.
But I don't think it's particularly good for the small web.
My Gemini capsule and gopher hole are NOT databases.
They are a collection of blog posts, terrible jokes, homelab
notes, CGI experiments, tutorials, an imaginary 400+ days old
plant (I plan on watering it until it reaches 666 days - that's
why it's name is Beelzebub), and whatever else I felt like
putting online.
The whole point is that:
‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾
Some of it is useful. Some of it definitely isn't.
If you want to browse the small web you actually have to go there.
You need software like curl for gopher, LaGrange for gemini.
You connect to a server. You browse a menu. You read someone's
weird little page. Maybe you follow a link and discover another
weird little page.
That's how small communities work.
With an AI sitting in the middle, the experience becomes: "Give
me everything interesting on the small web and summarize it."
The content gets consumed, but the community doesn't gain
anything from it:
* nobody visits the capsule/gopher hole
* nobody follows the links
* nobody says hello
* nobody starts their own Gopher hole, Gemini capsule or both.
Why using AI isn't always the solution
‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾
There is also something that younger generations will probably
never understand about getting information before AI: the pain of
spending three hours on Google trying to find the one paragraph
you needed for an essay. You would open twenty tabs, read
articles you didn't understand, follow links down completely
unrelated rabbit holes, and eventually find the answer somewhere
buried in a forum post from 2001.
It was frustrating, but somewhere along the way you learned
things you weren't actually looking for. You discovered context,
conflicting opinions, obscure facts and completely unrelated
information that somehow stuck with you.
Today you can simply tell an AI, "Give me a 500-word essay
about X," and a few seconds later you have something that looks
finished. It's incredibly convenient, but I think something is
lost in the process. Not necessarily information that was
required for the assignment, but all the accidental information
you would have encountered while searching for it.
The rabbit holes are gone. The happy accidents are gone.
And maybe those useless little pieces of knowledge were never
useless after all.
Programming before AI
‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾
I remember when I first starting juggling with PHP in 2004-2005.
(HTM) The PHP manual was always opened
along with Stack Overflow, some random blog from 2013, and
probably a forum post (using phpBB) that hadn't been updated
since PHP 4.0 which was released in May 2000.
You learned by constantly looking things up.
Now you can tell an AI, "Write me a PHP function that does this,"
and it will happily produce one in seconds. That's incredibly
useful, especially when you're stuck, but it also removes a lot
of the exploration.
You don't necessarily learn that array_filter() exists.
Or that you can compact() a bunch of variables into an array.
Or array_merge() 2 or more arrays into 1 array.
By using AI you get the answer, but skip the journey that would
have taught you the language.
What should we do
‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾
We should:
* keep the weird menus
* keep the tiny servers
* keep the badly formatted Gemini pages
Most importantly, keep the humans involved.
‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾
The small web is small for a reason. Let's keep it that way.
(DIR) Back to my phlog