Managing My Finances with Claude and MCP
Managing My Finances with Claude and MCP
I was going to vibe code a personal finance automation solution but then Claude Co-work implemented it for me. This post covers how I went from a lapsed spreadsheet to a working system that pulls transactions from 10 different banks and credit cards statements, categorises them and outputs an interactive dashboard.
I’ve always kept on top of my bills. For many years I had a running spreadsheet with the regular monthly expenses, making sure these were all paid for from a separate account. Around 2022 I started recording each transaction and manually categorising them with the aim of seeing where my spending went and how I could improve. As a Product Manager I know data is important and how can I keep on top of things without it.
It started by taking me an evening a month but as time went on life got in the way. In 2025 I really slipped and I didn’t even start in 2026. By that point I had an idea to vibe code a solution or build one with Make.com. I had the whole thing worked out with ChatGPT: the PRD, how we were going to try both workflows and test the parsing. It was going to be a proper project.
Around the time I discovered Claude Cowork I had also upgraded to the 5x max package and was starting to see the benefits of the increased allowances. I take my daughter to swimming lessons on Fridays and Saturdays and as there is no viewing gallery I have to sit in the car. I used the Claude app on my phone to ask about taking the entries from 10 different banks and credit cards and whether it was possible to do in Co-Work. This was the process that ChatGPT had told me would be hugely difficult months ago and had led me to putting it off. Claude said no problem.
So in 45 minutes in the car we sketched out the workflow, selected products and weighed up the pros and cons of Notion, Excel, Google Docs and Airtable against using a Claude skill or Make.com. Once I was done I asked for the summary, picked up my daughter and drove home.
Once at home I opened up Claude Cowork and got started. First I wanted to use the Google Drive MCP server so I could drop the statements in there and have Claude pick them up. This initially went really slowly. Despite connecting, I didn’t realise that the MCP only allows Claude to access Google Docs, Sheets and similar native formats, not arbitrary documents like PDFs. Because Claude just reported back that it couldn’t find the PDF I spent ages working on the connection, assuming it was a permissions issue. Once I Googled and fully understood the limitation I dropped the Google Drive component and decided to use files and folders on my computer instead.
Next we spent time deciding on the spending categories. I had many of these spread across my source documents but had never spent the time aligning them and making sure they were all sensible, so there was a small job to get these all set up and to decide on the mappings. Claude did most of the heavy lifting here.
I had selected Airtable as the destination for all the transactions. Airtable has some good visualisation features as well as being a versatile database so I thought I could utilise the platform. Claude set up the initial base quickly and I created the bases for each year. However after importing about half of the transactions I received an email telling me I had hit my free plan limit.
This isn’t Airtable’s fault. They have a good product that they need to charge for, and $20 a month is not bad. But it’s more than I want to pay for my personal finances. I’ve also found my SaaS spend has increased generally. I just took on a £90 Claude plan and I don’t want to spend more. Currently I’m looking to cut subscriptions, not take on new ones.
So I decided not to move forward with Airtable and Claude suggested Excel. I find this funny because for the last year I’ve been wondering why people are still using Microsoft products, and here I am with my Claude-generated AI solution building an Excel file.
I should be clear: I am not advocating for or suggesting that anyone use Microsoft’s cloud offerings. But I have an Office licence for my laptop and I’m happy to use the programmes locally. I also toyed with using Notion but rejected it over Excel. All I need is my finances tracked and to have Claude help me read them. So I’m using a spreadsheet to do exactly what a spreadsheet was designed to do. The licence was a one-off sunk cost so it’s costing me nothing extra and I can have Claude custmise the dashboard as needed.
The next step was to have Claude analyse my documents. I only decided to go back through the beginning of 2025, as the data before this time was solid, and the hardest part was getting the documents out of my different banks and onto my laptop hard drive. I started Claude off analysing and as more statements were added Claude picked them up and worked through them before I could give it the go ahead. Pretty soon it had all of my transactions.
I still need to work through them and make sure the categorisations are right. Some of the figures have been wrong so I’m working with Claude to go back and forth and fix them. But for now having an idea of totals is better than nothing and more important in the short term than granular accuracy.
Before I had even finished writing the paragraph above, Claude had built me a dashboard using HTML and Chart.js. It added a JSON export to the rebuild script so every time I regenerate the spreadsheet it also outputs a dashboard data file with all my spending pre-calculated. The interactive HTML dashboard shows KPI cards, category breakdowns, monthly and source charts, top merchants and a searchable transaction table. I just reload it after each spreadsheet rebuild to see the latest data.
I nearly vibe coded a solution, then I nearly paid for one and I didn’t need to do either. I’m learning that Claude can now be the main interface to all of my apps instead of defaulting to Make, Zapier or n8n, and I can probably step back from SaaS by using Claude and Office which, let’s not forget, powered companies for the past 40 years. I touched on this in my earlier post on what I’m building in 2026 and this finance project is one of those plans taking shape.
When building automations and solutions, picking the right products and making the right choices is much more important than using AI. Airtable is not a bad product. I know it’s good, I like it and I will absolutely tell my clients to use it where it’s necessary. It’s just not right for a small personal process when I’m in the middle of automating my own workflows and trying to reduce costs.
If you have similar processes sitting in spreadsheets that have fallen behind, or ideas for personal finance automation that feel too complex to start, the tools to do this are already there. The patterns I’ve built are reusable and the specifics change but the approach stays consistent.
