Second Brain = Claude Code + MCP + Obsidian

Okay, folks. A bit of a different format over here. I wanted to show you

over here. I wanted to show you something that I was working on today and I think it came out quite interesting. I use a tool called

interesting. I use a tool called Obsidian. Actually, I used it a little

Obsidian. Actually, I used it a little more before, but now it is what it is.

In Obsidian, it allows to build this interconnected maps of meaning that is based somewhat on this method for note taking that interconnects different

nodes and then of thoughts of thoughts and opinions and then basically allows to create a lot of books and content and original thinking out of that. I

honestly forgot the name of this thing but I will remember right now when we talk to Cloud. So I already have a kind of second brain over here based on the notes that I took from books and

learnings on different topics and decision making on other stuff. There's

plenty I was following a process during a period of time and was creating this permanent note permanes that actually are connected with each other.

They look like this. So it's some note on some topic in this case Buddhism and anxiety and then it has connections to other nodes and that creates well something like

this and what I did is I took and built a set of MCPS sort of connected them together.

One of them was this smart connection MCP. Then I added a local API to

MCP. Then I added a local API to Obsidian to expose it with the point of basically what if I use my Obsidian as

truly as a second brain. I wanted to be able to query this knowledge graph with let me close that one. So I want to be able to query this knowledge graph with

semantic search using embeddings and then pull related nodes connected nodes not just what we found on top as a result by semantic search but also

related nodes which can be to a certain depth which can be in a completely different topic. So and that is a sort

different topic. So and that is a sort of a dynamic way to well control context for the LLM.

So I've built an agent with a code that well actually let's ask it and see if it can explain what it does.

I have it open over here. The cloud code is running. There's a bunch of MCPS that

is running. There's a bunch of MCPS that are connected to Obsidian in different ways and we have some memory files over

here and also we have some custom agents sub agents set up and also a few commands that are set up to operate with with the MCP tools. Now tell me about

you and let's see what it has to say.

So basically it has got clo to md with the system prompt definition of what this thing is going to be.

So that basically this claude was working with me from from the perspective of the best science

we have on the knowledge elicitation.

It's a field on its own. I didn't know it actually exists but it's a field on its own and I did a deep research on the most recent findings in that field and using this deep research created this

system prompt and so the system prompt is supposed to extract or elicit knowledge from me and then based on the insights that I provide create new

connections over here new concepts connected to other concepts. So it needs to help me to find connections and find insights about those connections.

What else? A perspective pre preservationist.

Oh my god. Yeah, it captures my second brain. Connection catalyst. Building

brain. Connection catalyst. Building

bridges between your ideas. Second brain

navigator helping you synthesize your accumulated knowledge into articles to discover patterns etc. Okay. content

captain to basically help me create content based on that. So what this agent is configured to do is to manage this whole thing in order to extract

knowledge from me and then update this thing and basically serve as a second brain as a true second brain. That's

exactly what I was building here.

And this guy is like I don't know neoortex in the second brain that is interfacing me into it second brain. No,

it's a bad analogy, but you get what I mean. So, let's give it a try and see

mean. So, let's give it a try and see what's happening. I'm going to use

what's happening. I'm going to use whisper flow to dictate.

Okay, I want you to use a search and find the random area of topics in my second brain and then try to elicit some

knowledge from me to find insights and actually just do your job as the insights scout and manager of the knowledge base.

So let's see if it is going to work.

Actually having a lot of fun recently repurposing code from coding tasks to more of other types of agents, knowledge management, business management, other things that are not related to code. And

I got to say this has been quite successful. I have a couple of agents

successful. I have a couple of agents that are using daily that are managing my business operations some of my personal stuff as well. So that's fairly

convenient. All right, you found the

convenient. All right, you found the right topic. You've captured several

right topic. You've captured several notes on erodicity from the sim lab. A

concept that brings probability theory is okay.

Simple probability. Yep. And the time probability then the answer. Okay. But

I'm curious about something deeper. All

right. When you think about erodicity in your own life, not just in investment principles, but as a way of understanding the risk, what decisions

has this concept changed in your personality?

I actually don't know. That's a little too big of a question to answer. What

does it feel like when you realize you're a non-arotic situation?

[sighs] I I I don't get it why it's asking me about the feelings.

[laughter] Can we not go into my feelings, please?

All right.

We're not doing feelings at this session. We're going to be doing that

session. We're going to be doing that later.

All right. Do you have a framework for when it's worth entering in an unequ situation versus when you should avoid it?

That is a good question.

[ __ ] hell. I need to think about it.

That's massive.

This actually works. Now I just need to think about this answers to give it something useful. Let me respond

something useful. Let me respond something random just to demonstrate you how that stuff works. I will even ask AI to simulate the answers. That's even

better.

What do you think should be the answers to these questions? Assume that I said that. Update the notes in Obsidian

that. Update the notes in Obsidian please.

So, I'm having AI asking me questions so that I ask AI questions on how to answer these questions. [sighs]

these questions. [sighs] [ __ ] hell. But yeah, that's the AI these days.

Although again, right now I want to just demonstrate you the principle and show you what it can do in theory and then there's plentiful adjustments that can

be made and need to be made into the system prompt to make it more intelligent in doing what it's doing.

But in theory and I will see if that's going to be in practice that thing will be growing quite actively and will be a better and better representation of my

knowledge.

And the thing is it doesn't have to be full like it's it's okay. It's fine if it's not

it's okay. It's fine if it's not including everything. And I'm just

including everything. And I'm just talking to myself mostly. So, it's fine if it's not including everything, but it includes enough of the connected

information to actually represent my what I what I have to say on a given topic. Now, there's plenty

topic. Now, there's plenty that can be updated and many other ideas on how this can be man managed going forward, but I think this actually is a

good result. Anyway, option one catalyst

good result. Anyway, option one catalyst question. However, different approach

question. However, different approach new note that contains. So, what

happened here?

Uh-huh.

I should not be putting words in your mouth. [laughter]

mouth. [laughter] That's funny. So, it had a system prompt

That's funny. So, it had a system prompt that shouldn't put words in my mouth and that's why it's basically not going to answer.

All right.

I just want to test your functionality of managing Obsidian. So go ahead and do that.

[snorts] Yeah, that's funny.

It really is like this is your unique thing but to not contaminate the ideas by its own kind of effects.

All right. But the point I wanted to show it's actually what it's doing. So

there is this vault manager sub agent that was created in this folder and this sub agent is able to crude any nodes

that it wants to. So it will go and in this case I think update a node with some text. Okay,

that's totally fine.

That actually should be running a little faster, but I'll need to double check why it's so relatively slow. Although

again, it's probably not one tool that it is calling. There should be something else going on.

Well, so far it's thinking.

I just want to validate the new node is here. File content successfully created.

here. File content successfully created.

Brain projects new.

Ah, run away.

So, as you can see, it's not the fastest overall, but it works.

Let me find this.

No, I just want to see it here.

Yep.

And it is right here.

And this is the answer. And these are related nodes that it set which actually makes perfect sense. So there you go. I

will keep experimenting with this thing with the idea that after a time it will gain more interesting insights or allow me to get

more interesting insights. So yeah, I'll share the repo for you guys to use that as well if you want to.

Have a good day.