> ## Documentation Index
> Fetch the complete documentation index at: https://docs.footstep.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# parse_address

> Clean and structure messy free-text addresses

Correct and structure messy address strings. Fixes typos, expands abbreviations (`st` → `Street`, `ave` → `Avenue`), and infers missing fields like postcode, region, or country. Use this as a pre-step before geocoding when your input data is dirty; pass clean addresses straight to `geocode` or `batch_geocode`.

## Example prompts

* "Clean up these 30 customer addresses before I geocode them"
* "What's the proper formatted version of '10 downing st london uk'?"
* "Parse and standardise this list of addresses, then tell me which ones are in the UK"
* "Fix the typos in these addresses and break them into structured fields"

## What you get back

An array of results aligned with the input addresses. Each entry contains:

* `corrected` — the standardised, fully formatted address.
* `components` — structured fields (`house_number`, `unit`, `road`, `neighbourhood`, `locality`, `county`, `region`, `postcode`, `country`, `country_code`). Only populated fields are included.
* `corrections` — itemised list of changes made, each with `field`, `original` (or `null` if inferred), `corrected`, and `type` (one of `typo`, `abbreviation`, `inferred`, `reformatted`, `removed`). Omitted entirely when the input was already clean.
* `confidence` — score between 0 and 1. Omitted when 1.0; present when the parser was uncertain.

When every row resolves to the same country (or region), those fields are hoisted to a top-level `context` object instead of being repeated on each row. Mixed-country batches keep the per-row admin fields.

## When to use this vs `geocode`

* Your input is **clean** (real addresses, complete and well-formed) — skip `parse_address` and call `geocode` / `batch_geocode` directly.
* Your input is **messy** (typos, abbreviations, missing fields, inconsistent capitalisation) — call `parse_address` first, then geocode the `corrected` string. Cleaner input gives better geocoding hit rates and lets you flag low-confidence rows before spending geocoding budget on them.

## Limits

Up to 100 addresses per call.