MCP server extension to allow agentic access to moneydance data
Hi all,
This is my first post, forgiveness in advance if I'm not doing things the "right" way!
As a Claude chat / cowork user, I wanted a way for the AI agent to access my moneydance data so i can interact with it using natural langage. To that end, I created an MCP extension and a simple skill, which I have open sourced and you can find here (https://github.com/Spotlight250/md-mcp/releases/)
Obviously as an unofficially signed extension you will get warnings, so you use at your own risk - however as all the code is open to see in the repo you can see how it works yourself and hopefully be confident it is doing what I say it does!
Because it runs locally, it is private and secure - no data goes over the internet. However, obviously depending on how you use this, you will be sending data to the LLM and that may not be private depending on the provider and their privacy policy. Only use it if you are comfortable with that.
So far, I have found it incredibly powerful. I have used the subscription-finder skill to build a subscription dashboard and analyse price creep, and I have been talking to AI about my ISA portfolio risk profile and how I can flex this.
I would like to extend further, we shall see what happens time permitting.
For now, if you are interested take a look. If you wish to submit any improvements just raise a PR.
Thanks for listening!
Matt
Keyboard shortcuts
Generic
| ? | Show this help |
|---|---|
| ESC | Blurs the current field |
Comment Form
| r | Focus the comment reply box |
|---|---|
| ^ + ↩ | Submit the comment |
You can use Command ⌘ instead of Control ^ on Mac
1 Posted by Stuart Beesley ... on 15 May, 2026 11:56 AM
I have to say this is pretty cool. I was wondering what you were doing and how it worked. For anyone interested, this is my rough take of what it does.
At a simple level, this extension creates / runs a mini http web server and then listens to commands that are sent to it. You configure AI with knowledge of how to send commands to the local http server. The extension has a set of tools / features coded within it to know how to get the relevant data. So when AI asks for networth (for example), the relevant command is sent from AI via the node Java script bridge to the server that’s been run by the extension. The extension receives the command and then executes that feature. To add data extracts to the extension, additional features would have to be added to the extension.
Matt’s GitHub documentation covers it quite comprehensively.
I’m not commenting on the security of this here. But as a concept it’s pretty neat and I will be playing with it. It’s got me wondering about other uses.