A few days ago I shared AU Agent Utilities, a small open-source MCP server I built for deterministic Australian business-day calculations.
Since then I've been thinking less about the tool itself and more about the reason it needs to exist.
A good example is this:
At first glance, an AI agent could easily assume the answer is Monday 27 December.
But in NSW:
So the next business day was:
Wednesday 29 December 2021
That is exactly the kind of calculation I don't want an AI model guessing from memory.
LLMs are useful when a task requires interpretation, reasoning or language.
But some tasks have a correct answer based on a defined set of rules.
Business-day calculations are one of them.
If an agent is calculating:
then getting the date wrong by even one day can matter.
Instead of asking the model:
"What is one NSW business day after 24 December 2021?"
the agent can call a deterministic tool.
The MCP currently has four tools:
check_australian_business_day
Checks whether a date is a business day in an Australian state or territory and returns the previous and next business day.
calculate_australian_business_date
Adds or subtracts business days from a date.
count_australian_business_days
Counts business days between two dates with explicit start/end inclusion rules.
calculate_australian_deadline
Calculates calendar-day or business-day deadlines and can roll a result to the next or previous business day.
It currently supports:
ACT, NSW, NT, QLD, SA, TAS, VIC and WA.
The server runs remotely using Streamable HTTP.
MCP endpoint:
text
https://au-business-day.auagentutilities.workers.dev/mcp