Last week I saw someone give Claude “God Mode” with their Monarch Money account and thought—could I do that for Tiller?
Two days later: built my first Model Context Protocol. Claude now has direct access to my Tiller Money data (gsheet).
My approach:
Everything runs locally
Explicit tool permissions
Open source and auditable
I have no idea if this is the best way to do it, but it works for me. And honestly, that’s the point of sharing—we learn by building in public and seeing what others come up with.
Built this in a couple days with Claude Code.
Full repo and How-To: github .com/jackstein21/tiller-mcp-server
Would love to hear from other Tiller users who try it out or have ideas for improvements!
I think it’s a great idea if you’re wise with what you share. Account numbers, etc. I wouldn’t share. I prefer to do screenshots of my data into gemini or claude and have it analyze whatever subset I want it to see. I do have a google sheets script that goes out and pulls in the day’s market analysis based upon my holdings and that’s been helpful.
If you’re willing to trust it, sounds like it’s good for you.
Personally, I will not trust the hallucination machine with anything. I have used Gemini infrequently just to ask it how to do certain Sheets formulas if I have a particular solution I’m looking for and even that I trust as far as I can throw it.
sure AI hallucinate, but depending on the version of AI the issues are generally mitigated; it’s also similar to how people are building ad hoc reports for what they need; if they are familiar with the AI program and it can generate what they need faster all to them. For better or worse most of the time i’ve tried certain AI functions unless i know to piece meal it to it it’s not helpful.
Insightful input here , thank you all thus far. I’m so cagey and also a minimal user of AI relatively even though I know if someone or something wants after your data, they can get to it there are so many avenues.
But I agree on the learning when building in public! Thanks for leading the way and sharing here @jackstein21
I guess to add on; I’m assuming that most of us here are on non-enterprise versions of software. Of which a portion of our data then is used for some sort of “AI” training. ie with google’s smart feature that’s on by default: https://www.malwarebytes.com/blog/news/2025/11/gmail-is-reading-your-emails-and-attachments-to-train-its-ai-unless-you-turn-it-off
does read your data and give you features.
The main difference is that with the current AI chatbots and agents we’re also doing the tuning and debugging of the output vs with smart feature the developers have done more output testing before it’s released.
I’m not saying upload all your data to Claude; but depending where you are in life, test it is probably a better idea than not. ie what kevandcan does.
@jackstein21 I haven’t tried it and I probably won’t have time to set it up, but can you share more about what it’s doing for you? What problem is it really helping you solve?
Hey @heather ! Sorry for the delayed reply - to answer your question directly, I’m not a Tiller power user. I mostly just want to see where my money goes each month. The tool i made lets me have natural language conversations about my spending without having to navigate the spreadsheet or build formulas.
Really happy to see so many people chime in on this thread. It’s been cool to hear different perspectives on AI and personal finance.
My main use cases:
Thinking out loud about my finances with something that actually responds. It’s a good way to work through what I’m actually trying to get out of this dedicated tracking.
Planning ahead - “if I keep spending like this on dining out, where does that put me in 6 months?”
Using it as a friction point. Before an impulse purchase I’ll sometimes ask “how much have I already spent on X this month?” That 30 second pause is often enough to talk myself out of it.
I also think there’s value here for people who don’t have someone in their lives to give them straight feedback on their spending. That’s not my situation, but it’s a use case I had in mind.
On the privacy concerns raised earlier in this thread - totally valid questions. A few things worth noting:
The server runs 100% locally on your machine. Your data never leaves your computer.
It’s read-only by design. Can’t modify your spreadsheet.
All the code is open source if you want to inspect it.
You control your own Google OAuth credentials.
It’s not for everyone, but if you’re curious about conversational finance tracking and comfortable with a bit of technical setup, I think it’s worth trying. Happy to help anyone who wants to give it a shot.
Great explanation and use case Jack @jackstein21. I really like that the AI runs locally. I don’t trust all these AI tools being integrated into the core ecosystems. It’s starting to get to the point where it is a required core piece of functionality. There is a lot of data harvesting going on.
I can’t see everyone going thru the steps to setup the server component themselves. Like you said, “it’s not for everyone”. I would trust Tiller if they could implement the server component somehow and the users own gdrive. I’m not sure if that is even doable. I think having the ability to ask the type of budget feedback questions is valuable. It is sort of like having your own financial adviser available to get input on spending impacts on your budget and goals. I would think @peter has some done some deep thinking on AI and it’s use with Tiller. Peter, what do you think about this interesting use case and technology components?
