How to Parse US Addresses in Python

What the API returns

Every response is wrapped in a standard envelope. For parsing, the Result field is a single object whose properties are the address components. This is a real response for 1600 Pennsylvania Ave NW Apt 4B:

{
  "Id": "2737f8f3-af83-4ba1-b9c3-e29d2ba9e03b",
  "Result": {
    "inputAddress": "1600 Pennsylvania Ave NW Apt 4B Washington DC 20500",
    "addressLine1": "1600 PENNSYLVANIA AVE NW",
    "addressLine2": null,
    "addressNumber": "1600",
    "streetPreDir": "",
    "streetName": "PENNSYLVANIA",
    "streetPostType": "AVE",
    "streetPostDir": "NW",
    "unitType": "apt",
    "unitNumber": "4b",
    "city": "WASHINGTON",
    "stateCode": "DC",
    "zipCode": "20500",
    "zip4": null,
    "county": null,
    "matchMode": "Speculative",
    "matchTier": "Near",
    "confidence": 0.7,
    "matchCode": {
      "houseNumber": "Matched",
      "street": "Matched",
      "unit": "Matched",
      "city": "Matched",
      "state": "Matched",
      "zipCode": "Matched",
      "zip4": "NotApplicable"
    },
    "footnotes": ["recovered: standardized via correction, not an exact match"]
  },
  "ClientSessionId": null,
  "StatusCode": 200,
  "IsError": false,
  "Errors": []
}

The fields you'll use most often are covered in the component map below. Note that some fields are empty or null when they don't apply - there's no leading directional here, so streetPreDir is an empty string, and the parser didn't append a zip4 or county, so those are null.

Get your API key

1. Sign up at sthan.io and subscribe to the free Address Parser tier
2. Open your dashboard and create an API key
3. Copy the key - it looks like sthan_live_xxxxxxxxxxxxxxxx

You get the key immediately, with no approval queue. An API key is the simplest way to authenticate: you send it as a Bearer token on every request and there is no separate login step. (If you prefer a short-lived token, there is a JWT flow covered later.)

Configure the project

Install requests and keep your key out of source control by reading it from the environment:

pip install requests

# Set the key in your shell (or a .env file loaded by python-dotenv)
export STHAN_API_KEY="sthan_live_xxxxxxxxxxxxxxxx"

import os

STHAN_BASE_URL = "https://api.sthan.io"
STHAN_API_KEY = os.environ["STHAN_API_KEY"]  # raises early if it isn't set

Build the parser client

Use a single requests.Session so the connection pool and the auth header are reused across calls. The client URL-encodes the address, calls the endpoint, unwraps the envelope, and returns the Result object:

import requests
from urllib.parse import quote

# One Session for the whole app: connection pooling + a default auth header
session = requests.Session()
session.headers.update({"Authorization": f"Bearer {STHAN_API_KEY}"})


class ParserError(Exception):
    """Raised when the API reports a business error in the envelope."""


def parse_address(address: str, mode: str = "speculative",
                  timeout: float = 10.0) -> dict:
    encoded = quote(address.strip(), safe="")
    url = f"{STHAN_BASE_URL}/v2/address-parser/usa/{encoded}"

    response = session.get(url, params={"match": mode}, timeout=timeout)
    response.raise_for_status()  # surfaces 4xx/5xx as an exception

    envelope = response.json()
    if envelope.get("IsError"):
        raise ParserError(", ".join(envelope.get("Errors", [])))

    return envelope["Result"]

That's the whole integration. One call:

result = parse_address("1600 pennsylvania ave nw apt 4b washington dc 20500")

print(result["addressNumber"])   # 1600
print(result["streetName"])      # PENNSYLVANIA
print(result["streetPostType"])  # AVE
print(result["unitType"], result["unitNumber"])  # apt 4b
print(result["city"], result["stateCode"], result["zipCode"])  # WASHINGTON DC 20500

Choose a match mode

The match parameter controls how much typo tolerance the parser applies while standardizing components. The same call supports four modes, from strictest to loosest:

If you omit match, the endpoint defaults to speculative for the widest recovery. Whichever mode you pick, the location-defining parts of the address - the primary number, ordinal, directional, state, and the street's core name - are never substituted. A looser mode only widens tolerance for misspellings of the same street.

Map the components to your fields

Each component comes back as its own field, so mapping them to database columns is a direct copy. The fields you'll use most:

A few more components cover edge cases: streetPreType, streetPreMod, streetPreSep, and streetPostMod capture pre/post modifiers in unusual street names (for example "Avenue of the Americas"). They're empty for ordinary addresses. Pair the parse with matchCode - a per-component breakdown (Matched / Corrected / Inferred / Unmatched / NotApplicable) - so you can tell which fields were trusted as-is versus corrected. Here is a parse going straight into structured columns:

result = parse_address(raw_address)

record = {
    "address_number": result["addressNumber"],
    "street_pre_dir": result["streetPreDir"],
    "street_name":    result["streetName"],
    "street_type":    result["streetPostType"],
    "street_post_dir": result["streetPostDir"],
    "unit_type":      result["unitType"],
    "unit_number":    result["unitNumber"],
    "city":           result["city"],
    "state":          result["stateCode"],
    "zip_code":       result["zipCode"],
    "zip4":           result["zip4"],
}
# record is now ready to INSERT — one column per component

Alternative: JWT authentication

An API key is the simplest option and is all most apps need. If your security policy prefers short-lived credentials, the platform also supports a 2-step JWT flow. You call GET /Auth/Token once with your profileName and profilePassword headers, receive a token valid for up to 60 minutes, then send that token as the Bearer value on subsequent calls:

def get_token() -> str:
    response = session.get(
        f"{STHAN_BASE_URL}/Auth/Token",
        headers={
            "profileName": os.environ["STHAN_PROFILE_NAME"],
            "profilePassword": os.environ["STHAN_PROFILE_PASSWORD"],
        },
        timeout=10,
    )
    response.raise_for_status()
    return response.json()["Result"]["access_token"]


# Swap the static key for a cached token, refreshing on expiry
session.headers.update({"Authorization": f"Bearer {get_token()}"})

Everything else - the endpoint, the envelope, the parsing - stays exactly the same. Cache the token and refresh it shortly before the 60-minute expiry rather than fetching one per request.

Handle errors

Two status codes are worth handling explicitly so a hiccup never stalls a batch job:

• 401 - The key or token was rejected. Check the value and, on the JWT flow, refresh and retry once.
• 429 - Rate limit reached. Back off and retry rather than dropping the row.

import time


def parse_with_retry(address: str, mode: str = "speculative",
                     retries: int = 2) -> dict | None:
    for attempt in range(retries + 1):
        try:
            return parse_address(address, mode)
        except requests.HTTPError as exc:
            if exc.response is not None and exc.response.status_code == 429 \
                    and attempt < retries:
                time.sleep(2 ** attempt)  # 1s, then 2s
                continue
            raise
    return None

The exponential back-off (1s, then 2s) is enough for transient limits. For a large backfill, add a small delay between rows and a circuit breaker so one bad minute doesn't stall the whole queue.

What's next: confirm the parsed address is deliverable

Parsing gives you clean, structured components fast. It does not, on its own, confirm that mail or a package will actually arrive - a well-formed address can still point at a unit that no longer accepts delivery. When deliverability matters, run the address through the Address Verification API, which returns a deliverability status and appends ZIP+4 and county. It's the same envelope and the same requests pattern - one GET against /v2/address-verification/usa/{address}. The Python walkthrough is here: Verify US Addresses in Python.

If you want users to enter a clean address in the first place, Address Autocomplete suggests complete addresses as they type. For the full Python walkthrough covering parsing, verification, and geocoding together, see Integrate Address APIs in Python.