Claude MCP Tool That Cuts Token Costs by 99% — jCodeMunch

0:00 Hey kids, it's your old uncle Jay and I'm not using 11 Labs this time. This actually is it's me talking in the microphone thing. It's uh it's gotten to a point with the Parkinson's where actually using 11 Labs is even harder than talking. So, I'm going to talk if that's okay with you. Hey gang, a lot of you been around for a couple of years and you remember two, three years ago when uh we came out with a thing. What was it? Autogra. Autogro. The cool thing about autograph was it was an agent that could create other agents and you and I were all amazed by it, but the rest of the world didn't pay much attention. Well, this year they've come out with Open Claw and everybody loves it because Open Claw is an agent that can write other agents. So, you and I were like two years ahead of the curve there and I just wanted to congratulate you for

0:46 that. Oh, and then later on about a year later, I came out with the uh with the thing that uh where our agents could talk to each other like they had their own Facebook. I forgot what it was called, but then uh this year Molt Book came out and everybody was like, "Oh, Moltbook is so cool." So, you and I have been years ahead of this thing all along. And we're about to be years ahead again as I present to you. Where is it here? Uh oh, Jode Munch, my my latest latest thing here. And um like it says here, you you can save like nearly 100%. I know you can't save 100% of your tokens because you'd be getting nothing done, but you can save like 99% of your tokens on a lot of your calls using Jode Munch. This is another one of those things where two three years from now

1:33 they'll somebody will have this built in to the LLMs, but for right now, it's it's just you and me. So, JCode Munch, you'll find it out at GitHub there. All you got to do is do the git cloning thing. And what is it's an MCP server. I forget what MCP stands for. Model model model context. I don't know. I'll cut this I'll cut this part out later and make myself sound smarter. Um, but you can see here all I did was I just cloned it down to a to an empty uh MCP's directory. And installing it is pretty easy. I've got it installed on all my cloud instances both here at home and at work. And if I do a listing of the MCP servers here, you'll see JCode Munch is installed. What JCode Munch really does in a nutshell is um it's kind of like if

2:20 you need to look something up in the encyclopedia currently Claude like takes every volume of the encyclopedia and reads it and you're like, "Hey Claude, I just asked you about pomegranates. All you needed was the P volume." And by the way, if I'm losing you kids, uh ask your grandparents what the Encyclopedia Bratannica was. In fact, go ahead, ask them what books were because you're probably not using those a lot anyway. Um, what you could do is you can you could just tell Claude to uh install the MCP and it'll do it. You don't have to worry about the big long read me in the in the repo. You you can do it if you want. This here is just uh this part is just explaining to you what I did with the what I did better with the uh encyclopedia analogy. But um bottom line, you install and run this bad boy

3:07 in Claude and you're going to get a lot more bang for your buck and for what they're charging for tokens over there at Anthropic. It totally makes it worth it. By the way, it's not just Claude specific. any any uh anything you have that uh that supports MCPs can use JCode munch MCP and you can start saving tons of tokens um right away. The uh primary benefits of using it are spelled out here if you want to if you want to hit pause and read that. I don't like reading things to people online. You know that about me. I like people read their own things. That way you can fast forward if you want. Oh, here this is cool. This is where I asked it to run a quick fundamental. Now I'm reading to you like I said I wouldn't but uh run a quick fundamental test of demonstrating the token savings and Claude will go

3:52 ahead and it'll it'll just write a short little demonstration program and it'll spit out what the token savings are and it's doing that right here. It's going to show you some stuff. Nobody really knows what any of this stuff means. So just if somebody's watching you just look at it and nod and look smart. But uh you just got to wait for Claude to get done thinking and then then boom, here it is. Here it is telling you that instead of reading the whole file, it's just going to read what it what it has indexed. Um, oh, speaking of indexing, uh, Claude will, uh, also as you make code changes will will reindex the code so it keeps working as you, as you work on your programs. I'm making this sound a lot more complicated than it is really. Here's the bottom line. Install JCode MunchMCP. Start saving mad tokens

4:41 with your uh, with your LLM programming. Here's the uh the actual test results from uh from the test that Claude wrote. And you can see here that what am I looking at? Oh yeah, here here I'm telling it, okay, you went and installed everything. Search for something specific and contrast both the with the Jcode Munch MCP and show us what the difference is so that you, my faithful and loyal audience, will be all uh all impressed with with the JCode Munch here. as I hope you are. I know it's been a while since I made one of these videos. Uh my health hasn't been all that good. So, I appreciate you being able to sit on your hands for a couple of years while while I get my crap together. Uh well, my crap is together

5:27 now. I have a note from my doctor and everything. And uh here you'll see the the uh Jcode Munch is running with the MCP and it's grabbing the symbols that it is indexed. And this it's all a lot of words to say. What this shows you the bottom line bottom line is that it used 700 tokens whereas if you didn't have the MCP server would have used 3,850 tokens. So that's that's more and you want less. So this does it 5.5 times fewer tokens for this single lookup in this. But don't don't take my word for any of this. Go ahead and and download the MCP server. Um install it and run it against your own test. Run it against some of your own code. run it. You can pull down a repo if you want and try it

6:13 against that. You can try it against something that you've got written on on your machine. Um, it's going to work and it's going to save you tokens. And tokens is going to save you money. It'll save you time, too, because obviously it takes a lot less time to process 700 tokens worth of information than 3,850 tokens worth of information. I got this uh site. I'll I'll link to this. This is out of my gllus us site, but this is a breakdown of uh of all the details with the MCP server. If you're the geeky kind who likes to drill down into things. I'm the geeky kind who likes to show you the stuff you can drill down into. But between you and me, if if I was looking at this, I would I probably wouldn't read most. But um yeah, this demonstrates the old way of doing things versus J code Munch. And uh that's why I

7:01 titled it that, I guess. But look at look at the contrast there. 141,000 tokens versus 480. Another 99.7% savings. And um here's the actual data I used from a I had a I had like a project from work that I ran this against for the stuff on this page. And uh yeah, it just it just it just always uh it always fascinates me when I ask it. Hey, hey, man, how much did we just save doing that? Plus, you're going to see that if you if you're the kind like I am, I monitor what my claud usage is throughout the process because gez, I asked I asked one query what what opus I had a question on opus cost me $6 to answer and that's when I got to work on this on on this thing on indexing the

7:47 code and uh it's just it's just paid off in volumes. Here's showing you real dollar impact just just based on this one program. If if I'm working on this per year, I'm saving like 15 grand in in token costs. Um, not that I'd be writing that same program all year, but programs like it, if I were doing that sort of work day after day, which I do, um, that's what I'd be saving in this case, what I'd be saving my boss. So, there's the wrap up on there. I I hope you'll download it and thanks again for listening.