I think n8n is a good starting point. Claude is great, TONS of power. Just talk o it and it does great things. However it will often miss details. I've seen it store passwords in clear text files, in JSON files, and I have to correct it. If I didn't understand I wouldn't have caught it and there things would sit insecure. With n8n you have to piece it together, build it. And most of the time no AI needed, no tokens spent, nothing You can even get a free tier Google cloud VM to run it on. With n8n you will read API doc's, learn how it works in the background. Get to know and understand it. This will make a world of difference in how you interact with Claude. You will have more precise instructions. You will know where to correct it. If you don't understand the process you just ask Claude for something and the lack of clarity means it has to fill in the blanks. Can it do that? Sure it can but it might not be exactly what you want. Think of buying a house. You get a realtor. They set a search because you aid you want a 3 bedroom house under $500k. That's it. Think of the results. Now to say I want a 3 bedroom, 2 1/2 bath, single story, house. No condos or townhomes, I want it in this reached, No older that 25 years, Impact windows (I I've in FL so that's a big deal), Not in a flood zone, 1+ or better car garage. Now what results will you get. The better you understand something the better you will be at knowing the details. And it will make you a better Claude user. The stuff people churn out using Claude but have zero understanding of anything is slop. AI in the hands of somebody who understands the process elevates the way beyond those that don't understand. We are in the early phases of AI. Ive been in IT a long tike and I an guarantee at some point the people churning out slop will fall away because those who use Claude as a tool to elevate their work will e the ones who rise to the top and stand out. When AI drops a feature because they deemed it obsolete those ppl will know how to adjust.

I've been reading and looking into, testing using the Model Workspace Protocol (MWP) methodology with my Claude projects. If this will move things around and cleanup memory files I'm curious if that will total break that setup?

I just went through this and it's pretty cool. I even added some tools and changed the LLM to use a local Ollama model that worked great. So I'm not paying for LLM API calls. I am already working through in my head how I can expand this and add different input methods so it will work with more than one source, Slack, WhatsApp, maybe Siri?

Nice video. Thanks for putting this together. One issue I keep running into with Claude Code is how it handles secrets. It tends to store API keys in JSON files, which are plain text. In one case, I had to explicitly tell it to create a .gitignore and exclude those files so they would not get pushed to GitHub. From a security perspective, this is a problem. My day job revolves around securing credentials and enforcing rotation policies. Storing high-privilege keys in flat files goes against basic security hygiene. If a key grants access to modify workflows or automate actions, the risk increases. An exposed key could allow someone to inject malicious logic, alter automation steps, or introduce code that looks harmless at first glance. Most people will not manually inspect every workflow change at a deep level. I approach systems with an attacker mindset. I ask how I would compromise this if I had access to it. Plain text secrets are low-hanging fruit. Once access is gained, the blast radius can grow quickly. For anyone using Claude Code with n8n or similar tooling, I strongly recommend: - Use environment variables or a proper secret manager - Keep secrets out of source control - Enforce key rotation - Review generated project structures before committing - ensure you are using .gitgnore to help prevent accidental publishing of sensitive files Automation is powerful. Treat credentials like production assets, not temporary strings in a JSON file.

I agree. n8n is not going anywhere. People keep acting like there is room for one tool. That idea fails in practice. n8n and Claude Code solve different problems. They work better together than in isolation. You grow faster when you understand multiple tools and their limits. You make better decisions when you know which tool fits a task and which one does not. Teams benefit from people who can switch approaches without friction. Focusing on a single tool creates blind spots. Blind spots slow you down. They also make it harder to pivot when requirements change. Engineers who cannot adapt fall behind quickly.