Prompting AI Agents

Users normally interact with this server through an LLM or AI agent, not by typing raw MCP calls. The user asks an economic question in ordinary language, the agent chooses one or more read-only tools, and the server returns structured metadata, series, and observations.

This page describes the recommended routing patterns. It is not a hard guarantee that every model or MCP client will choose the exact same sequence. The server provides tool descriptions, schemas, resources, and prompt templates; the LLM client decides when to call them.

Main routing patterns

Use ordinary economic language, but include the indicator, geography, frequency, date window, and preferred output format when those details matter.

User intent	Typical MCP route
Ordinary analyst request for GDP, CPI, unemployment, cash rate, credit, exchange rates, or yields	list_economic_concepts then get_economic_series
Transparent formula-based indicator such as real cash rate, yield-curve slope, or credit-to-GDP	get_derived_series
Exact ABS, RBA, or APRA source control	search_datasets or list_catalogue, then get_abs_data, get_rba_table, or get_apra_data
Exploratory or quick-turnaround request	describe_dataset, get_latest_observations, or get_top_observations

Prompt examples

Quarterly real GDP growth

User prompt:

List the quarterly real GDP growth data for the past 10 years.

Typical tool calls:

list_economic_concepts(query="quarterly real GDP growth")
get_economic_series(concept="gdp_growth", last_n=40)

Why: the prompt describes a curated economic concept and a quarterly 10-year window. Forty quarterly observations is the compact way to express that window when the user wants the past 10 years of available data.

Latest cash rate target

User prompt:

What is the latest RBA cash rate target?

Typical tool calls:

list_economic_concepts(query="cash rate")
get_economic_series(concept="cash_rate_target", last_n=1)

Why: the cash rate target is a curated semantic concept, and last_n=1 asks for the latest available observation.

Real cash rate

User prompt:

Compare the real cash rate over the last year.

Typical tool call:

get_derived_series(concept="real_cash_rate", last_n=12)

Why: the real cash rate is part of the server’s transparent derived-series layer. The response includes formula and operand provenance in metadata.derived.

Housing credit source discovery

User prompt:

Find the best ABS or RBA data source for housing credit.

Typical tool calls:

list_economic_concepts(query="housing credit")
search_datasets(query="housing credit")

Why: the first call checks whether a curated semantic shortcut already exists. The second call ranks source-level ABS, RBA, and APRA catalogue entries for cases where the user needs a dataset or table recommendation before retrieval.

Prompting tips

• Ask for the output format you need, such as a table, a short summary, or a comparison.
• Include frequency and window language when relevant, such as monthly, quarterly, latest, last 12 observations, or 2020 to 2024.
• Ask the agent to preserve source identifiers or provenance when you need reproducibility.
• For exact ABS/RBA/APRA work, name the dataflow, table, publication, or series ID if you already know it.
• If a result looks ambiguous, ask the agent which MCP tool it called and which metadata fields identify the resolved source.

For direct call syntax, see Examples and the Tools reference.