Claude Persistent Memory: Vibe Marketing Unlock

Key Takeaways

What we're looking at is a Claude project that I use to manage my ongoing subscriptions. Since Claude can now scan email, I'm having it look for emails related to renewals, trials, subscriptions, and then cataloging those items as a ledger for me to keep track. However, that's not the reason I wanted to make this video. I wanted to make this video because I discovered an amazing unlock that allows Claude to have persistent memory as I was creating this project. Let me show you it in action. Ongoing. I would like to have services that are pay as you go categorized as ondemand, not active. So I'm changing a fundamental way on how my project works and how it tracks a specific category of platforms. Let's see what happens. All right, the project has looked at the relevant services and identified a bunch and now it's recategorizing the status from active and it should be now on demand. Okay, look, they're popping up. Great. It's deleting the old records and replacing them with the new information. All right, but here's the magic ongoing. It's going to know now to classify this style of service as on demand because it's dynamically updating my project instructions and effectively this gives claude project a persistent memory that it can always reference as needed and within that document are instructions on how to maintain and update its own instructions. So let's look these are the instructions living in a Google document. It was last updated about 15 minutes ago. The incoming instructions will have a time stamp of 2130. When this time stamp changes, we'll know that our instructions have been updated. All right. And there it is. The instructions have been dynamically updated. When I told Claw that I wanted to make this ongoing, that is a trigger word in the instructions that gives it permission to make these updates. So, how does this work? Let's look at how the project is set up. If you're learning about Vibe marketing, it's these kinds of tactics that are going to separate the good from the great. Understanding how to leverage capabilities like MCP servers are what's going to unlock the creativity for how you show up, whether it's building systems or processes for yourself or for a client. Understanding how these things work is critical. So, if nothing else, this is a great opportunity to get your hands dirty creating a custom MCP tool. If you like this video, please like and subscribe. And if you want a guide for this, I'm happy to create one. Just drop me a comment and I'll turn one around in a couple of days. I encourage you to sign up for the viable edge newsletter where I'm sharing strategies and tactics like this and a lot more. All right, let me show you this thing. In my project instructions, I'm referencing a specific document, the Claude project platform ledger instructions, which is in my project knowledge right here. And it is the document that we have pulled up. So effectively, we're pointing the instructions to a document in project knowledge and we're using N8N via the MCP trigger to dynamically update this project in an N8N workflow. So let's see how that works. For this project, I have an MCP trigger. And that MCP trigger is visible in the cloud chat box. When we open up our tools menu, the platform ledger. Within the platform ledger, you can see I have several tools here. Each one of these tools align with one of the items that are connected to the MCP server trigger. So these are all the tools that are reflected here. And in our update ledger instructions, this tool is calling another workflow. And here's what this workflow looks like. So effectively what's happening here is that once Claude wants to update the instructions, it's retrieving the existing instructions with this node and then it is replacing them with the updated instructions coming in with the data payload in the MCP trigger call. Now, if that sounds confusing, trust me, it's a lot easier to set up than it sounds. You can ignore this code node here. I was using this earlier on when I was having formatting issues. Let me explain why that can be a challenge. The way that this Google doc node works, the second one to update a document works based on find and replace or you can insert content after a specific point because we want these instructions to be living and breathing and we don't want to keep expanding it with updates. We want to find and replace the entirety of the content every time. Therefore, the incoming string has to be character for character match of the entire instruction document. Fortunately, the string of text that is coming through from the retrieval enough to be recognized by the find and replace. So hopefully that's not too complicated. The whole point is that it's less complicated than you might imagine. So I've deactivated the code node for now in case any bugs pop up. Sometimes there's a character in the instructions that will trigger a JSON error, but I've updated it now several times without this code parser and it works. This is a powerful unlock for Claude. So, all you really need to make this work is Claude Desktop, N8N, and Google Docs. And if we think about the possibilities that this opens up, it's pretty incredible. We just walk through two-way street with the project instructions, right? We have Claude that can read the instructions via project knowledge and then write back into those instructions or project knowledge effectively giving it a persistent memory. But imagine that you had several cloud projects just like this and these projects were related to each other. For example, what if you're running a virtual marketing team or you're working with a client and you have different projects within that client engagement? Well, if you have one master project and you have sub projects that have specific tasks, like a content creation project, now your other projects or sub projects can have visibility into the persistent project knowledge that your primary project is constantly updating. So now if you have projects that are support projects or whatever you want to call them, they can have up-to-date ongoing context on whatever the main project is. And similarly, your main project can point to the project knowledge of your sub projects and also be persistently upto-date. So you can now imagine the possibilities that this opens up for managing cloud projects and maintaining persistent memory over time. And I can't wait to explore this further. I'm sure there are easier ways to do this when you start to get into vector databases and rag and all that kind of stuff. But for a dude with Claude Desktop, N8N, and Google Docs, this unlocks a new dimension of capability with Claude. So, let me break this down for you one more time so you can set this up for yourself. And if you want a guide, I'm happy to create one. Just leave a comment. I'll whip one up in a couple of days and I'll send it to anyone who wants it. So again, here's how it works. We have our project instructions stored in a Google doc. The name of the instructions matters because in our project instructions, we're directly referencing this exact file. Now, that file is included in our project knowledge right here. When we want to make updates, we're using an MCP server via N8N. You can see the various tools within my MCP server. And those tools align with what we see here, which have the various tools for managing my ledger as well as updating the project instructions. When Claude wants to update instructions, it'll go to the MCP server. It'll trigger this tool to then call the next workflow, update ledger instructions, which is this. When this tool is called, the incoming data payload will include the new instructions that Claude is updating. The first Google Docs node is capturing the existing instructions because we will need to identify the full string of text to find and replace. I'm not using this code node, but I'm keeping it there for safekeeping in case I run into errors. And then this Google Docs node is taking the new data coming in with the payload, finding the text that is coming from the previous Google Docs node and then replacing it with the data that's coming in from the MCP call. So, I hope that is clear. It's very easy to set up. Like I said, if you want instructions or a guide, I'm happy to create one. I'm really excited to play with this new capability and see what's possible by providing Claude the dynamic ability to update its own project knowledge and instructions. I really believe this is a huge unlock. I already have tons of things that I want to try out with this. And if you have any ideas, please share them. Leave a comment. If you like this video, please like and subscribe and sign up for the viableedge.com newsletter. There's a link below. Thank you for watching.