ausecon-mcp-server 0.6.0

MCP server for Australian economic data from the ABS and RBA.

ausecon-mcp-server

ausecon-mcp-server is a Python Model Context Protocol (MCP) server for structured Australian macroeconomic and financial data. It exposes a small, source-aware tool surface over Australian Bureau of Statistics (ABS) and Reserve Bank of Australia (RBA) datasets and returns provenance-rich JSON suitable for downstream analysis.

The server is intentionally narrow in scope. It does not try to be a general economic chatbot. Instead, it gives MCP clients reliable discovery and retrieval tools that can be composed into research, policy, and analytical workflows.

Status

Releases are published to PyPI and versioned via git tags. See CHANGELOG.md for release history.

The current release includes:

• curated discovery across expanded ABS and RBA catalogues
• deterministic search ranking across dataset IDs, aliases, names, tags, and descriptions
• active-only RBA table discovery by default, with optional inclusion of discontinued tables
• ABS dataset structure retrieval
• ABS data retrieval in a normalised response shape
• compatibility with current live ABS structure and CSV payload shapes
• RBA table listing and retrieval
• a semantic resolver whose curated default concepts now narrow to concrete series
• stricter validation for tool inputs and parameter ranges
• source-aware upstream and parse error messages
• retry logic for transient ABS and RBA upstream failures
• provider caching keyed to upstream requests rather than client-side filters
• support for RBA event-style tables such as a2

At this stage, the server should still be treated as an opinionated early release rather than a complete coverage layer for all ABS and RBA content.

Tool Surface

The MCP server currently exposes the following tools:

Tool	Purpose	Key inputs
search_datasets	Search the curated ABS and RBA catalogue with deterministic ranking	query, source
get_abs_dataset_structure	Retrieve ABS SDMX dimensions and code lists	dataflow_id
get_abs_data	Retrieve ABS data in a normalised response shape	dataflow_id, key, start_period, end_period, last_n, updated_after
list_rba_tables	List curated RBA statistical tables	category, include_discontinued
get_rba_table	Retrieve an RBA statistical table in a normalised response shape	table_id, series_ids, start_date, end_date, last_n
get_economic_series	Resolve a small set of high-value economic concepts to ABS or RBA retrievals	concept, variant, geography, frequency, start, end

Currently supported semantic concepts

get_economic_series currently supports:

• cash_rate_target
• headline_cpi
• trimmed_mean_inflation
• gdp_growth

By default, these currently resolve to the following narrowed series:

• cash_rate_target -> RBA a2 cash rate target series (ARBAMPCNCRT)
• headline_cpi -> ABS CPI all groups CPI index for Australia, quarterly (1.10001.10.50.Q)
• trimmed_mean_inflation -> RBA g1 year-ended trimmed mean inflation (GCPIOCPMTMYP)
• gdp_growth -> ABS ANA_AGG quarterly real GDP growth (M2.GPM.20.AUS.Q)

variant, geography, and frequency are validated against the catalogue entry. For ABS datasets, populated variants can be literal SDMX keys or partial fragments that are completed against the live structure. For RBA tables, populated variants narrow the response to the declared series IDs. Variants that are declared but not yet populated raise a clear "not yet wired" error.

Discovery And Validation

• search_datasets prioritises exact dataset or table IDs, then exact aliases, then exact names, then broader multi-term matches.
• common economist phrasing is normalised for ranking, including terms such as "jobless", "mortgage", "rates", and "fx".
• discontinued RBA tables are excluded from search_datasets by default.
• list_rba_tables excludes discontinued tables by default and returns a discontinued boolean field on every row.
• empty identifiers and empty search queries are rejected before any network call.
• last_n must be positive when provided.
• get_abs_data validates annual, quarterly, monthly, and half-yearly ABS period strings.
• get_rba_table validates ISO date bounds.
• get_economic_series validates start and end after resolving the target source.
• transient ABS and RBA upstream failures are retried automatically.
• malformed upstream payloads are surfaced as source-aware parse failures.
• search_datasets scores should be treated as ranking metadata rather than a stable contract.

Response Shape

Data retrieval tools return a normalised payload with three top-level sections:

• metadata: source, dataset or table identifier, retrieval URL, and related retrieval metadata
• series: long-form series descriptors including labels, units, frequency, and dimensions
• observations: long-form observations keyed by date, series_id, and value; some RBA observations may also include raw_value when the upstream cell is non-numeric

This design keeps source provenance explicit while making downstream processing simpler in Python, R, or other analytical environments.

Discovery Coverage

The curated catalogue is still intentionally selective, but v0.5.0 covers the main analyst workflows more credibly:

• ABS prices and inflation, labour, national accounts, activity, housing and construction, external sector, and lending indicators
• RBA monetary policy, payments, money and credit, interest rates and yields, exchange rates, inflation, output and labour, and external sector tables

Requirements

• Python 3.10+
• uv (for the uvx launcher used by every client below)

Installation

The server is published to PyPI and is intended to be launched by an MCP client on demand via uvx. No manual clone or install is required — the client configurations below handle everything.

To confirm the server is reachable on your machine, you can run it once by hand:

uvx ausecon-mcp-server

uvx will download the package into an isolated, cached environment the first time. The server speaks MCP over standard input/output and waits for a client to connect, so expect it to sit idle until an MCP client attaches.

Connecting An MCP Client

This repository currently provides a local stdio MCP server only. Claude API / Anthropic MCP connector setups and other remote HTTP-based MCP connectors require a separately hosted HTTP server, which is out of scope for this repository today.

Claude Desktop

An example Claude Desktop configuration is included at examples/claude_desktop_config.json:

{
  "mcpServers": {
    "ausecon": {
      "command": "uvx",
      "args": ["ausecon-mcp-server"]
    }
  }
}

Claude Code

Add the server with the Claude Code CLI:

claude mcp add --transport stdio ausecon -- uvx ausecon-mcp-server

Codex

Add the server with the Codex CLI:

codex mcp add ausecon -- uvx ausecon-mcp-server

Cursor

Add an entry to ~/.cursor/mcp.json (or the project-level .cursor/mcp.json):

{
  "mcpServers": {
    "ausecon": {
      "command": "uvx",
      "args": ["ausecon-mcp-server"]
    }
  }
}

VS Code

Add an entry to .vscode/mcp.json in your workspace (or the user-level equivalent):

{
  "servers": {
    "ausecon": {
      "type": "stdio",
      "command": "uvx",
      "args": ["ausecon-mcp-server"]
    }
  }
}

Resources

In addition to tools, the server exposes the curated catalogue as MCP resources. Clients can render these in a resource picker without calling search_datasets first.

URI	Returns
ausecon://catalogue	Flat index of every curated ABS and RBA entry (id, source, name, description, category, frequency, tags).
ausecon://abs/{dataflow_id}	Full curated catalogue entry for a single ABS dataflow (e.g. ausecon://abs/CPI).
ausecon://rba/{table_id}	Full curated catalogue entry for a single RBA statistical table (e.g. ausecon://rba/g1).

All resources are read-only, served as application/json, and sourced from the static curated catalogue — no network calls are made to render them.

Prompts

The server registers four prompt templates that chain the existing tools into common economist workflows. Clients such as Claude Desktop surface these as slash-commands.

Prompt	Arguments	What it does
summarise_latest_inflation	months: int = 12	Pulls headline and trimmed-mean CPI via get_economic_series and summarises them against the RBA 2–3% target band.
compare_cash_rate_to_cpi	start: str, end: str | None	Narrates the path of the cash rate target against headline CPI over the window.
macro_snapshot	as_of: str | None	Assembles a compact snapshot table of cash rate, headline CPI, trimmed-mean CPI, and real GDP growth.
discover_dataset	topic: str	Runs search_datasets and list_rba_tables for the topic, then recommends the top two candidates.

Each tool is also annotated with readOnlyHint and openWorldHint so compliant clients can indicate that the server only reads from external, evolving sources.

How To Use The Server

The most reliable workflow is:

1. Use search_datasets when you do not yet know the exact ABS dataset or RBA table.
2. Use get_abs_dataset_structure before get_abs_data when you need to inspect valid ABS dimensions and keys.
3. Use get_abs_data or get_rba_table for retrieval once you know the target dataset or table.
4. Use get_economic_series only for the small curated semantic shortcuts listed above.

Example client requests

In an MCP-enabled client, the user can ask for things such as:

• "Search for datasets related to trimmed mean inflation."
• "Show me the ABS structure for CPI."
• "Fetch the last 12 observations from RBA table g1."
• "Get headline CPI from 2023 onwards."
• "Get quarterly real GDP growth from 2020."

Example retrieval patterns

Discover relevant datasets:

search_datasets(query="cash rate")

Inspect active inflation tables and optionally include discontinued RBA coverage:

list_rba_tables(category="inflation", include_discontinued=True)

Inspect ABS dataset structure before querying:

get_abs_dataset_structure(dataflow_id="CPI")

Fetch a filtered ABS dataset:

get_abs_data(
  dataflow_id="CPI",
  key="all",
  start_period="2024-Q1",
  end_period="2024-Q4",
  last_n=4
)

Fetch an RBA table:

get_rba_table(
  table_id="g1",
  last_n=8
)

Fetch an event-style RBA policy table:

get_rba_table(
  table_id="a2",
  last_n=8
)

Resolve a curated economic concept:

get_economic_series(
  concept="cash_rate_target",
  start="2020-01-01"
)

Resolve trimmed mean inflation using the default narrowed series:

get_economic_series(
  concept="trimmed_mean_inflation",
  start="2020-01-01"
)

Resolve quarterly real GDP growth:

get_economic_series(
  concept="gdp_growth",
  start="2020-Q1"
)

Development

Install the development environment:

uv sync --python 3.12

Run the test suite:

uv run pytest

Run linting:

uv run ruff check src tests

Repository Structure

src/ausecon_mcp/
  catalogue/   Curated ABS and RBA discovery metadata
  parsers/     Source-specific parsers for ABS and RBA payloads
  providers/   HTTP retrieval and cache-aware source adapters
  cache.py     Simple in-memory TTL cache
  models.py    Shared normalised data structures
  server.py    FastMCP server entry point and tool registration
tests/
examples/

Releasing

If you want to publish a release from this repository:

1. ensure pyproject.toml contains the intended version
2. commit the release-ready state
3. create an annotated git tag such as vX.Y.Z
4. push the branch and the tag to GitHub
5. create a GitHub Release from that tag with release notes

An example tag command is:

git tag -a vX.Y.Z -m "vX.Y.Z"
git push origin main
git push origin vX.Y.Z

Once the tag is on GitHub, you can create the release in the GitHub interface under "Releases" -> "Draft a new release".