FXMacroData for Haystack

Build Haystack research pipelines with official economic observations, release calendars and searchable Documents that retain their source metadata.

Subscribe to FXMacroData for non-USD data, full available history, FX, commodities and positioning. Use the public USD workflow to evaluate the integration before connecting your subscription.

Explore FXMacroData · API documentation

The public USD catalogue, recent macro history and release calendar support evaluation without an API key. The history example requests the most recent 90 days. Data availability varies by series; your subscription and its terms govern protected access.

Install from source

With the companion client and this project in sibling directories:

python -m pip install ./fxmacrodata-public-client ./haystack-fxmacrodata

This command uses source packages, without assuming a package-registry publication. Requires Python 3.10+ and Haystack 3.1.1+.

Pipeline

from haystack import Pipeline
from fxmacrodata_haystack import FXMacroDataFetcher

pipeline = Pipeline()
pipeline.add_component("macro", FXMacroDataFetcher(operation="indicator_history"))
result = pipeline.run({"macro": {"arguments": {"currency": "USD", "indicator": "inflation"}}})
documents = result["macro"]["documents"]

Connect documents to document writers, retrievers, joiners or prompt builders. Each Document contains a complete record and source metadata. data retains the unmodified endpoint response; records provides a tabular view. The error output distinguishes an unsuccessful request from a valid empty result.

For Haystack Agents, use create_tools() to obtain 72 separate FXMacroDataTool objects with complete documented input schemas. For a smaller agent context, pass operations=["data_catalogue", "indicator_history", "release_calendar"]. See CAPABILITIES.md.

Run python examples/usd_macro_brief.py to build and execute a real three-component Pipeline without a language model. Run python examples/tool_invocation.py to call a native Tool directly.

Subscription credentials and persistence

The default is Haystack Secret.from_env_var("FXMACRODATA_API_KEY", strict=False). An absent variable keeps public USD access available. Set public_only=True to ignore ambient credentials. Use Secret.from_env_var with your application's own secret-variable name if desired. Environment references serialize into saved pipelines; credential values do not. Haystack deliberately refuses to serialize Secret.from_token values.

When loading a trusted saved pipeline, allow this specific extension module: Pipeline.loads(saved_yaml, allowed_modules=["fxmacrodata_haystack.integration"]). This preserves Haystack's normal deserialization protection.

Preserve units, timestamp flags, provenance, empty results and the distinction between FXMacroData-generated forecasts and market consensus. The plugin preserves MCP visual resources but does not render MCP Apps.

README links contain static referral parameters. This integration adds no analytics SDK, identifier or click beacon.

Test

python -m pytest tests -n 8 --dist load

Code is Apache-2.0 licensed. API access, data use and brand rights remain governed by their applicable terms.