rba-mcp

mcp-name: io.ausdata/rba-mcp

Ask Claude about Australian interest rates, exchange rates, and lending rates and get real, current numbers — not "I don't have access to that data." This MCP server gives Claude (and other MCP clients like Cursor) live access to the Reserve Bank of Australia's statistical tables, with curated mappings for the most-asked indicators.

Hosted access? For cross-source queries, webhooks, an always-on REST API, and a uniform response envelope across all 9 sources, see ausdata.io — free tier available (500 calls/mo, no card).

Companion to abs-mcp (ABS macro stats), ato-mcp (ATO tax + ACNC charity data), and au-weather-mcp (Australian weather via Open-Meteo + BOM) — together the four cover the most-asked Australian official data.

What you can ask

Once installed, your LLM can answer questions like:

Every answer comes with the period, units (Per cent per annum, USD per AUD, etc.), the publication date, and a link back to the RBA source. The MCP wraps RBA's CSV statistical tables and exposes them through 5 plain-English tools.

Install

# After publish:
uvx --upgrade rba-mcp

# Local dev:
uv pip install -e .

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "rba": {
      "command": "uvx",
      "args": ["--upgrade", "rba-mcp"]
    }
  }
}

Why --upgrade? uvx rba-mcp (without the flag) uses whatever wheel is cached and never adopts new PyPI releases on its own — Claude Desktop's MCP child process keeps running the same wheel until you fully quit the app and refresh the cache by hand. --upgrade makes uvx check PyPI on each launch and pull a newer release if one exists. Recommended for everyone except offline-first / pinned-version workflows. To verify which version is currently serving you, look at the server_version field on any DataResponse (added in 0.1.5).

If you also have abs-mcp installed, both servers run side-by-side. Claude disambiguates with the server prefix (rba:get_data vs abs:get_data).

For local dev (pre-PyPI):

{
  "mcpServers": {
    "rba": {
      "command": "uv",
      "args": ["run", "--directory", "/absolute/path/to/rba-mcp", "rba-mcp"]
    }
  }
}

Cursor

Add to ~/.cursor/mcp.json (or workspace .cursor/mcp.json):

{
  "mcpServers": {
    "rba": {
      "command": "uvx",
      "args": ["--upgrade", "rba-mcp"]
    }
  }
}

Tools

Curated F-tables

For these five, series accepts plain-English keys (e.g. "aud_usd" instead of "FXRUSD"):

• F1.1 — Money Market — Monthly: cash rate target, cash rate, bank bills, OIS rates, treasury notes
• F4 — Retail Deposit & Investment Rates: transaction accounts, savings, term deposits, cash management trusts
• F6 — Housing Lending Rates: owner-occupier vs investor, variable vs fixed, outstanding vs new loans
• F11 — Exchange Rates — Monthly History (1983+): AUD/USD, AUD/EUR, AUD/GBP, AUD/JPY, AUD/CNY, AUD/NZD, TWI
• F11.1 — Exchange Rates — Daily (2023+): same series, daily resolution

Any other F-table works too — pass raw RBA series IDs (e.g. "FXRUSD") instead of curated keys.

Worked examples

"What's the current RBA cash rate?"

latest(table_id="F1.1", series="cash_rate_target")

"AUD to USD over the last year"

get_data(table_id="F11.1", series="aud_usd", start_period="2024")

"Compare AUD against USD, EUR and GBP since 2020"

get_data(
  table_id="F11",
  series=["aud_usd", "aud_eur", "aud_gbp"],
  start_period="2020"
)

Period formats

RBA series use ISO-style date formats. Pass start_period / end_period (or the legacy start_date / end_date aliases) as:

start_period snaps to the first instant of its period; end_period snaps to the last. So start_period="2024", end_period="2024" returns "all of 2024", not just 1 January. The legacy start_date / end_date parameter names continue to work as aliases (rba-mcp <= 0.2.x); prefer the new names for cross-sister consistency with the rest of the portfolio.

Verifying your install

The running MCP server reports its version on every DataResponse:

{ ..., "server_version": "0.1.8", ... }

If you see a value below the latest on PyPI, your uvx cache is stale. Either switch to ["--upgrade", "rba-mcp"] in your config (recommended), or refresh manually:

uvx --refresh rba-mcp --help
# Then fully quit and relaunch Claude Desktop (Cmd+Q — window-close is not enough).

Claude Desktop's MCP child processes are long-lived; refreshing the wheel cache does not restart an already-running server. Cold app launch is required.

Development

git clone https://github.com/Bigred97/rba-mcp.git
cd rba-mcp
uv sync --extra dev
uv pip install -e .

# Unit tests (no network)
uv run pytest

# Live integration tests (hits RBA CDN)
uv run pytest -m live

The SQLite cache lives at ~/.rba-mcp/cache.db. Data refreshes every 6h, latest 15min. Delete to force a refresh.

How it works

Claude picks the right tool, fills in the curated series keys, calls the live RBA CDN, and synthesises the answer. When the curated table is stale relative to a rate decision (RBA's monthly F1.1 publishes around the 5th business day, but Board meetings can hike between publications), Claude fluidly composes web-search results with this server's data:

You don't have to know what FIRMMCRT or FXRUSD mean — and neither does Claude. The server's curated YAMLs map plain-English keys (cash_rate_target, aud_usd) to RBA series IDs and surface unit attribution + the CC-BY 4.0 attribution string in every response.

Sister MCPs (Australian Public Data portfolio)

The portfolio runs side-by-side in any MCP client; Claude disambiguates via the server prefix (rba:latest vs abs:latest vs ato:get_data vs weather:latest).

Want all 9 sources behind one REST API? The hosted gateway at ausdata.io adds cross-source joins, full history, webhooks, and HMAC-signed responses on top of these MCPs — free tier (500 calls/mo, no card).