Searching the forum with a chatbot: rembetiko.gr’s MCP (Claude, ChatGPT, etc.)

For the forum
1

Good morning, everyone!

The forum now has an MCP server. In practical terms, this means that you can connect an artificial intelligence chatbot, such as Claude or ChatGPT, to your account here. The chatbot can then search and read our discussions on your behalf and answer you with references to the relevant posts.

We have more than 26 years of discussions (the oldest posts date from 2000) and thousands of topics. A chatbot like this can dig up things that you would have a hard time finding on your own using the search function.

What is MCP?

MCP (Model Context Protocol) is an open standard. It allows artificial intelligence tools to connect to external services. The chatbot does not “know” on its own what has been written on the forum. Through MCP, however, it gains a set of “tools” that it can use, such as:

  • searching topics and posts, using Discourse’s familiar search syntax (before:, after:, order:oldest, #category, @user, etc.)
  • reading entire topics or specific posts
  • lists of the most popular or most recent topics, filtered by category or tag
  • member profiles and their activity
  • drafts: the chatbot can prepare a draft post for you, and you can read it, correct it and publish it yourself

The server address is:

https://rembetiko.gr/mcp

What the chatbot “sees”

The chatbot connects using your own account and sees exactly what you see. Nothing more. When you connect, the forum asks you to approve what permissions it will have (for example, read-only access, or permission to save drafts as well). You can revoke access whenever you like.

How to connect

Menus often change from one version to another, so treat these steps as a general guide.

Claude (claude.ai or the Claude Desktop app)

  1. Settings → Connectors → Add custom connector.
  2. Name: Rembetiko.gr, URL: https://rembetiko.gr/mcp.
  3. Click Connect. The forum page opens; sign in (if you are not already signed in) and approve access.
  4. In a new conversation with Claude, make sure the connector is active (using the tools button below the message field).

ChatGPT

  1. Settings → Apps & Connectors → Advanced settings, then enable Developer mode.
  2. Back in Connectors, click Create and enter the same URL.
  3. Sign in with your forum account and approve access.

There are other tools that support MCP as well (e.g. Claude Code, Cursor and VS Code). The process is similar: enter the URL and sign in.

Two real examples

Below are two investigations we conducted through MCP. I describe what was requested, what the chatbot did and what it found. All the results come from forum posts and include links to them.

1. “Find me the earliest discussions about the three-string and four-string bouzouki”

Question: What are the oldest discussions on the forum about the three-string bouzouki? I want to see how the “three-string or four-string” debate began.

The chatbot searched for τρίχορδο order:oldest, that is, posts mentioning it, from oldest to newest. It then read the most interesting ones. Here is what it found:

  • 27 February 2000: the first relevant question, in the topic Bouzouki or tzoura?. A beginner who plays baglamas is thinking of getting a three-string instrument and is hesitating between a tzoura and a bouzouki.
  • November 2000: the topic Bouzouki Instruction Book. This is where the question “Three-string or four-string bouzouki?” appears in full for the first time.
  • October 2001: the half-joking, half-serious Detetrachordize the bouzoukia!, from a supposed “Radical Movement for Rebetiko Dechiotification”. Aris writes a long and very substantial reply there. He points out that “the damn four-string bouzouki wasn’t invented by Chiotis. It had existed for a very long time” and that what Chiotis added was “guitar-style tuning”. As early as 2001, he was already complaining that “new friends don’t take the trouble to ‘leaf through’ the older postings a little”. So here is a good reason for us to do exactly that!
  • May 2002: in the topic Makam - Byzantine Modes, frank mentions that the oldest known photograph featuring a bouzouki is of Nikos Milanos (Kolatsis) in Portaria, Pelion, from 1892. He also writes that, according to Milanos’s grandchildren, he played “a regular, three-string bouzouki, with wooden tuning pegs, from around 1880”.

All of this came from one search and a few readings, in a minute or two.

2. “Study a discussion and summarise it for me”: the Piraeus mode

Question: What has been said on the forum about the Piraeus mode? How does it differ from Hitzazkiar, and which songs are written in it?

The chatbot located the topic Piraeus mode: 57 posts from 2018 to 2023, with contributions from liga_rosa, pepe, alk, icetelli, Nikos Politis, Peloponnisios and others. It read the entire topic and produced the following summary:

  • Structure: it basically has the intervals of Hitzazkiar, with a permanently (or almost permanently) raised fourth degree, which returns to its natural position only in the final phrase. The melody also revolves around the fifth degree rather than moving from the octave down towards the tonic (liga_rosa, #10 and #50).
  • Correspondence with a makam: there is no clear correspondence. Zirgkioulelî Souzinak with a permanent sharp on the fourth (Voulgaris), Sirf Hitzazkiar, and Tsiamoulis’s definition (#19–#24) have all been mentioned. Pepe notes that the Piraeus mode is probably the only case in which manes are named after a folk mode rather than a makam. He has identified 12 recordings titled “Piraeus” that are genuinely in this mode (#49).
  • Tunings: mainly the Syros tuning (D–G–B♭), as well as D–B–G and D–B♭–B♭ (#20).
  • Songs mentioned: O Memetis, O Bochoris, O Xemagkas, My eyes have seen many things, Your enchanted eyes, At Manthos’s tekke, At Perdikakis’s tekke, Karantouzeni, We were smoking one evening, and others.
  • The most interesting discovery: in 2023, pepe wonders whether the Piraeus mode is “the only mode that the people of Piraeus did not cultivate”, since the classic examples are from Asia Minor (#37, #39). Peloponnisios replies with songs by Markos and Gogos (#40). Nikos Politis opens up the discussion of what “Piraeus school” ultimately means (#42).

If the chatbot tells you something interesting, ask it for the link and read the original post. The summary is a starting point, not a substitute for the discussion.

Other ideas for research

  • “Find all discussions about Skarvelis before 2010 and tell me which topics keep recurring.”
  • “What different versions of the lyrics have been discussed for this particular song?”
  • “I want to buy a second-hand bouzouki. Gather the advice given on the forum about what I should look out for.”
  • “What has member X written about tunings?”
  • “What are the most popular topics of all time in the Instrument Making category?”
  • “Is there already a topic about what I want to ask?” (Very useful before opening a new topic!)

A few tips

  • Search is keyword-based, not meaning-based. If it cannot find something, ask the chatbot to try variations: with and without accents, “hitzaskiar”/“hitzazkiar”, “Tsitsanis”/“Tsitsani”, or even Greeklish for the older posts.
  • Always ask for links. A good chatbot will refer you to the specific post. If something sounds odd, go and check it.
  • The chatbot can make mistakes. It may misinterpret a post, confuse who said what, or “fill in the gaps” with its own knowledge. Check anything based on the forum. Treat anything without a link with caution.

Give it a try and tell us here what you found, what worked and what did not. Especially if you dig up some forgotten gem from the old days!

3 Likes
2

Well done, @chrispanag.

I think this is a fantastic addition because it gives anyone with knowledge of AI the ability to search the forum in a different way. This means they’ll be able to find valuable information that would otherwise be missed by ordinary searches, and it will also be a great help to researchers of rebetiko!

The forum is being upgraded and moving forward!

3 Likes
3

You know how wary I am of AI, both for ideological reasons (addiction to convenience, the erosion of critical thinking) and ecological ones (the enormous waste of energy and drinking water).

If someone wants to find out more, perhaps the latest and most up-to-date posts, combined with the forum’s search function, would be enough. It might prove useful for a possible mapping of the forum’s thirty-year history.

Christos, were your latest posts replying to some threads that had been started without receiving a response written with the help of AI?

4

Blunting critical thinking? I don’t know. It certainly helps you study things a lot, and the fact that you have to have an idea and scrutinise it sceptically doesn’t exactly contribute to blunting it, because left to its own devices, it’ll make a complete mess.

1 Like
5

Niko, there are different ways to use AI!

I now use it every day at work and increasingly in the rest of my life too: to organise my finances (guess who does my tax return — I don’t need an accountant!), book appointments and reply to pointless bureaucratic emails. The other day, I told it to contact an e-shop’s customer support so that I could return a product.

Now I can focus on the things that matter, rather than fiddly tasks that simply eat up my time!

I’d wanted for a long time to be able to reply on the forum, especially to new members’ posts, with topics that had been discussed in the past and were relevant to what they had posted about. Unfortunately, the most time-consuming part of this is searching through the forum so that the information provided is genuinely useful. AI helped me with that. I asked Claude to read some unanswered posts and find three or four related topics on the forum that might be of interest. After two or three minutes, it had given me this list for each one.

Some weren’t particularly relevant, so I removed them after reading them. Most of them, however, were to the point. I then wrote the messages. :slight_smile: So, in reality, the part where AI helped was retrieving relevant information — which is also what takes the most time.

1 Like
7

Nikos, I understand your reservations. Those of us involved in learning see it every day: when someone else thinks on your behalf, you don’t learn. That’s what happens with an answer key, and it’s what happens with AI too.

I only slightly disagree with you on the word „ideological“.
What you’re describing involves questions that can be tested through research, and researchers have started doing so. For example, in an experiment with high-school students studying maths, those who worked with plain ChatGPT solved the practice exercises better, but performed worse in the test without assistance. Some other students used a version of ChatGPT that functioned like a teacher, giving hints rather than a ready-made solution. They did not suffer the same loss (Bastani et al., PNAS 2025). In other words, the issue is whether you use the tool instead of doing your own thinking.

Christos described exactly this: the chatbot brought up the relevant threads, he read them, discarded those that didn’t fit, and then wrote it himself. What it saved him was the searching, not the thinking. On a 26-year-old forum, where valuable posts get buried under thousands of newer ones, that matters! Of course, it requires suspicion, as @ZARAZ says too. AI hallucinates answers, so anything it tells us needs a link and cross-checking.

You’re right that there is an environmental cost. I haven’t looked into how large it is per query, and I see that estimates vary widely. If anyone has a reliable source, it would be interesting if they could share it here, because the news and media often exaggerate this kind of information!

Perhaps I’m wrong too… :slight_smile:

3 Likes
8

Regarding the environment specifically, I’ve spoken to people who work at the major AI data centres and have learned a great deal about the subject, as it concerns me too.

  1. Yes, data centres use a great deal of energy, but most of the time they don’t put a strain on the grid — for the most part, they rely on renewable energy sources (solar, or even more so geothermal!). In fact, because they usually produce more energy than they consume, they sell the surplus energy to the grid, often replacing “dirty” energy sources such as lignite.
  2. The claim about drinking water is a myth. Yes, they use some liquid for cooling, but it is a closed-loop system. They don’t draw water from somewhere and then throw it away. It’s like someone filling a large Olympic-size swimming pool once, and then reusing that same water. They don’t discard it and draw in more. If you compare it with the water requirements of a football stadium, or even worse, a golf course, the differences are astronomical (the data centre uses far less water over time).
1 Like
9

It depends on the data centre, the energy sources used to generate electricity in that particular country, how often UPS generators are used because of problems with the grid, and what fuel/technology they use. There’s no simple answer.

I understand that “generative AI” can be a useful upgrade to search, but if it starts filling the forum with AI-written walls of text, that would be the end of the forum for me. It would be a thousand times better to have us tearing each other apart over the four-stringers and the skyladika crowd than this.

2 Likes
10

I think you’re right, @emc. I believe that, at the moment, AI is useful for facilitating certain tasks that would otherwise be difficult for someone to carry out without a machine. In any other case, it would be problematic.

That’s an interesting thought, and it hadn’t crossed my mind at all!
I’m trying to put myself in the shoes of someone arriving for the first time looking for some information, as well as someone who has information to share. The person searching will come, have a look and leave—or at most ask a question or two. But why shouldn’t someone who has information share it if they have something to say? Is it because their knowledge is limited, or because they feel they have very little to contribute compared with the others?
Let’s hear what everyone else thinks about this too!

1 Like
11

For me, it depends! If you simply say, “We’ve discussed this before here,” you shut down the conversation and don’t give the new member a chance to share their own perspective. But if you point out where it was discussed, mention some of the previous participants and add a few open questions that would be interesting to resolve, then you open up a new round of dialogue.

2 Likes
12

That’s right, I’m terribly bored of reading discussions I take part in (or don’t even take part in).

What’s more, since I’m bored of music too, I send the little bot to good live gigs to listen on my behalf.

If I go out myself and then have to drive, I get the little bot to drink the raki for me.

Seriously now: if it really is a myth, it’s a very widespread one. I’d like to learn more: is there a concise source that presents the actual ecological cost of using artificial intelligence?

1 Like
13

There are uses and uses… You can set it to write and reply on your behalf — but you can also ask it: “In this 2,000-message thread, is there anything related to Tsitsanis’s wife?”

There are certainly situations where this could prove useful, and I don’t think it’s laziness not to want to read 2,000 messages to find a needle in a haystack…

14

And come on, tell me…
when our brain is “working”, don’t we consume energy?
What do you think grey matter is?

2 Likes
15

And since I was curious, it turns out I wouldn’t have found this result otherwise.

image
image1920×950 181 KB

1 Like
16

No, on the contrary: we produce. The mind is the perpetual-motion machine. It produces without expending anything.

1 Like
17

After days away due to other obligations:

Regarding the waste of energy and water, one of the many relevant articles containing specific data is this one:

I agree with Stelios and Periklis. The less we delegate to AI, the better it is for our physical and mental wellbeing. And of course, a forum with AI-generated content would be uninteresting and have nothing to offer.

That’s a major trap. I wouldn’t trust a language model with matters that serious (medical matters too), nor would I want my accountant to lose his job.

I accept that as far as searching is concerned; it gives me a hard time too whenever I look for an old discussion. But it’s also a process that sharpens the mind: figuring out how to search more effectively, or remembering phrases from the discussions.

Why do replies to orphaned messages remind me of the classic AI style?

18

I’d put it slightly differently: the human mind is extremely expensive and consumes resources whether it produces anything or not. With the appropriate use of AI, it can produce more—or generate more output. However, bureaucratic matters require extra care; I’ve caught it not knowing the latest decisions and laws, and when it comes to scientific matters, it tends not to know the latest research.

19

If you keep asking and challenging him, eventually he’ll either find what he couldn’t find before or at least admit that he can’t find it, so you’ll have a better idea of where you stand.

3 Likes