Automation and AI
AI integration (MCP)
I connect AI assistants like Claude and ChatGPT to the services you already use, so you can ask them about your own data and, if you allow it, have them act on it. I have written MCP servers for Firefly III and Midas, and in both the permissions are split so the assistant cannot do more than you want.
What you get
- An MCP server for your service
- MCP lets an assistant read and work with a service you use. If the service has an API, I write the server that connects it.
- Separate read, write and delete permissions
- The assistant can only do what you allow; writing and deleting can stay off.
- Your approval before important actions
- Actions such as sending an order can wait for your confirmation; in mcp-midas the order tools are off by default.
- Runs with your own keys
- The connection uses your service's address and your own token, with no hosted service of mine in between.
- Claude, ChatGPT and agents
- MCP is an open standard, so the same server can be used from Claude, ChatGPT or an agent such as Claude Code.
- A setup note
- How to connect it, which permissions are on and how to switch them off, written down plainly.
How it goes
- 01
You tell me the service and the questions
Which service, and what you want to ask the assistant or have it do.
- 02
We set the permissions
What the assistant may read, what it may change, what stays closed, and the price.
- 03
I write and test it
On a test account or real examples first, with writing switched off until it behaves.
- 04
Hand-over and note
Connected to your assistant, with the note; if the service changes, you can reach me.
What the time depends on. It depends on how large the service's API is and how many operations you want the assistant to reach. A read-only connection is the quickest start, and writing can be opened later.
Work I have done
Questions
What is MCP?
Model Context Protocol: a standard way for AI assistants like Claude to connect to a service, read from it and, if you allow it, act on it.
Can the assistant do anything it wants?
No. Reading, writing and deleting are separate permissions; anything you have not allowed is simply not available to it.
Does my data go to the AI company?
What the assistant reads to answer you is sent to the AI service you use, under that service's terms. That is why the permissions matter; you decide which data it can see at all.
Tell me about it
A few quick choices, then your details. It takes a minute, and I write back by email.
Step 1 of 3
mcp-firefly-iii