I connected Claude to Jira at work and it works great. I am NOT a jira fan so Claude handling it for me is extremely valuable to me. Same for GitHub. I've never written such good and detailed PRs since I handed it over to Claude.
I use the Consensus connector for my academic work and I love it. I have scheduled a task for looking for new academic papers in my field every month. I also use it for starting a literature review.
Hi Ruben, I’m new to your newsletter but really great stuff, thank you. I run a small healthcare clinic and for obvious reasons worried about data security - especially using Claude and especially using Connectors as we have sensitive info in our Microsoft ecosystem and other apps. Are there any good resources you would recommend to understand this area more?
Healthcare on Connectors is the exact scenario worth thinking hard about. The model layer is well-defended, but the execution layer (where agents actually read your inbox and act) is where the gaps live. I wrote about this in https://thesynthesisai.substack.com/p/the-wrong-war. Practically: start with read-only scopes, audit which connector actions touch PHI, and check whether your BAA with Microsoft covers agent-mediated access. Most don't yet.
The important part is using this carefully. Connect the tools that actually improve your work, not everything. More access should always mean more responsibility.
You are right! I am using Claude for many Agentic AI, Security Research, and tools jobs. Each time, we have to be very careful in crafting a clear, concise prompt; otherwise, it is just a waste of a token with no productive answers, only the same repetitive code lines.
Thanks, man, for saving me, too. Let me try these hacks and get this done in a cost-effective way. The monthly package Claude just ran away within some time, 10 to 15 min.
What an extremely timely post! I was trying to figure out how to custom-connect my StoryGraph to Claude last night, but I found it a bit complicated. Then you posted this, and Section V really simplified things. Thanks for sharing, Ruben!
🎯Thanks for sharing! This Hands-on guidance is Exactly what we Clauders need to explore our personal ”3D Latent Space ” of what’s possible, desirable and doable in our daily interaction with Claude. Personally, I’m on my first 120 days with Claude Max in dozens of Projects and multiple dozens of threads-inside of which I have developed user-operated interactive innovation models I need to learn how to propagate. Claude Max is amazing at 3D Instantiation-did you know that?
Modules to enable building and instantiation of an agentic LLM-on-a-chip adapted Vertical Model enabling instant sovereign Human-in-the-loop semi-automated Innovation/invention , based on systematic LLM Latent Space exploration.
🎯”very technical” . Indeed.So also in the opinion of Claude who disapproved of my effort to be specific, and instead suggested ”I build the coordinate system that AI uses to explore ideas systematically rather than producing the same answer over and over.” 🤓
It's basically an AI-powered insight management tool where you can build your own knowledge system and use AI to make sense of everything you've collected :)
The three-connector starting point is right. I went straight to connecting everything and spent the first week working out what Claude was pulling from where instead of actually using it.
Meeting notes, email, team messages is a clean boundary because those are the three places where working context actually lives. The thing I want to see is whether connectors hold state between sessions or re-pull fresh each time. That distinction matters for whether it works as memory or just as a fast lookup, and those are different products.
Just joined your paid subscription and can't access the Slack channel! Looking forward to the new link. Thanks so much for your amazing work - learning a ton from you.
Thank you so much. I love your info. However, I'm trying to customize based on my needs. I'm an author who wants to promote his novels by posting memes on various social media channels using Buffer. I'm trying to use AI to automate this task but so far am coming up dry. (I don't see a way to connect Buffer to Claude.) If I put memes in Notion, could Claude automate this task for me? Thanks.
With the move of managing connectors to customize- is selecting “needs approval” the same as turning off as you mention above? I only see disconnect as the only other option.
I connected Claude to Jira at work and it works great. I am NOT a jira fan so Claude handling it for me is extremely valuable to me. Same for GitHub. I've never written such good and detailed PRs since I handed it over to Claude.
haven’t tried it so this is really nice to know - which GitHub connector did you use exactly?
We actually have our own GitHub enterprise instance so we have an MCP connector
that’s super cool :)
I use the Consensus connector for my academic work and I love it. I have scheduled a task for looking for new academic papers in my field every month. I also use it for starting a literature review.
i ran the same thing with Zapier for arXiv - saved me hours back then :)
This is a great newsletter! Keep sharing your AI knowledge 🙏
that’s nice to know :) my newsletter will stay free
and let me know if there’s any preference for a future topic
Hi Ruben, I’m new to your newsletter but really great stuff, thank you. I run a small healthcare clinic and for obvious reasons worried about data security - especially using Claude and especially using Connectors as we have sensitive info in our Microsoft ecosystem and other apps. Are there any good resources you would recommend to understand this area more?
probably this one from Anthropic itself: https://support.claude.com/en/articles/11175166-get-started-with-custom-connectors-using-remote-mcp
Healthcare on Connectors is the exact scenario worth thinking hard about. The model layer is well-defended, but the execution layer (where agents actually read your inbox and act) is where the gaps live. I wrote about this in https://thesynthesisai.substack.com/p/the-wrong-war. Practically: start with read-only scopes, audit which connector actions touch PHI, and check whether your BAA with Microsoft covers agent-mediated access. Most don't yet.
The important part is using this carefully. Connect the tools that actually improve your work, not everything. More access should always mean more responsibility.
100% read-only first :) otherwise Claude will do random things you didn’t ask for
You are right! I am using Claude for many Agentic AI, Security Research, and tools jobs. Each time, we have to be very careful in crafting a clear, concise prompt; otherwise, it is just a waste of a token with no productive answers, only the same repetitive code lines.
i hit those limits too fast before that i wrote this newsletter: https://ruben.substack.com/p/how-to-stop-hitting-claude-usage
i still do these hacks :)
Thanks, man, for saving me, too. Let me try these hacks and get this done in a cost-effective way. The monthly package Claude just ran away within some time, 10 to 15 min.
hope it helped a lot!!
It does.
What an extremely timely post! I was trying to figure out how to custom-connect my StoryGraph to Claude last night, but I found it a bit complicated. Then you posted this, and Section V really simplified things. Thanks for sharing, Ruben!
hey i’m glad this came perfectly - do let me know how it turned out :)
Yep! Will do 😊
#42
Granola + Gmail + Slack is the stack.
I set this up last month and I've opened maybe 3 tabs total since.
Everything lives in Claude now.
Meeting context, email threads, team messages.
This trio compounds fast.
we’ve been doing it before the newsletter came out :)
each tool alone is useful but all 3 together is a beast - Claude has full context on every info decided
Btw let’s connect!
Love it, simple and yet really powerful. Using a similar set up but with fathom instead of granola.
🎯Thanks for sharing! This Hands-on guidance is Exactly what we Clauders need to explore our personal ”3D Latent Space ” of what’s possible, desirable and doable in our daily interaction with Claude. Personally, I’m on my first 120 days with Claude Max in dozens of Projects and multiple dozens of threads-inside of which I have developed user-operated interactive innovation models I need to learn how to propagate. Claude Max is amazing at 3D Instantiation-did you know that?
dozens of Projects, you’re deep in it - what specifically are you making?
Modules to enable building and instantiation of an agentic LLM-on-a-chip adapted Vertical Model enabling instant sovereign Human-in-the-loop semi-automated Innovation/invention , based on systematic LLM Latent Space exploration.
that’s very technical aha all the best to you :)
🎯”very technical” . Indeed.So also in the opinion of Claude who disapproved of my effort to be specific, and instead suggested ”I build the coordinate system that AI uses to explore ideas systematically rather than producing the same answer over and over.” 🤓
That nice
thank you :)
Nice post, Ruben!
We actually just rolled out MCP support for Claude in Cluing as well, so this space is moving pretty fast.
that’s nice to know :)
What’s cluing btw? just curios
It's basically an AI-powered insight management tool where you can build your own knowledge system and use AI to make sense of everything you've collected :)
Claude really improved this year I hear haha
they got a surge of users earlier this year too
The three-connector starting point is right. I went straight to connecting everything and spent the first week working out what Claude was pulling from where instead of actually using it.
Meeting notes, email, team messages is a clean boundary because those are the three places where working context actually lives. The thing I want to see is whether connectors hold state between sessions or re-pull fresh each time. That distinction matters for whether it works as memory or just as a fast lookup, and those are different products.
they re-pull fresh every session, there’s no persistence
honestly it makes the connector choice more important :)
100% agree
Just joined your paid subscription and can't access the Slack channel! Looking forward to the new link. Thanks so much for your amazing work - learning a ton from you.
hey Jacqueline - we’re moving the community to Circle, and yes, i dm you the link!
Hi Ruben, greetings from South Australia. New subscriber chasing an updated Slack link. Cheers.
we’re now in Circle - sent you a dm for the invite link :)
Thank you so much. I love your info. However, I'm trying to customize based on my needs. I'm an author who wants to promote his novels by posting memes on various social media channels using Buffer. I'm trying to use AI to automate this task but so far am coming up dry. (I don't see a way to connect Buffer to Claude.) If I put memes in Notion, could Claude automate this task for me? Thanks.
Notion is a workspace + database tool - stores your memes, images, captions. it can’t schedule tasks for you if that’s what you want to do :)
i asked Grok, try this:
Log into Buffer → go to https://publish.buffer.com/settings/integrations
Find the Claude / MCP section → authorize and copy the MCP server URL it gives you.
Open Claude Desktop (or Claude Code) → Settings / Customize → Connectors → “Add custom connector” (or “Add remote MCP”).
Paste the URL → connect.
Wow. Thank you. I will see if I can get this to work. I assume I need the premium Claude (and using the desktop version), correct?
Pro plan and desktop app, yes :)
With the move of managing connectors to customize- is selecting “needs approval” the same as turning off as you mention above? I only see disconnect as the only other option.
not the same,
1/ needs approval: the connector is still fully active and available
2/ disconnect: the only option that actually turns the connector off