streetworks

An open Python SDK for UK street works APIs — one consistent, typed, well-tested client for the services the sector actually uses.

We do this not because it is easy, but because it is hard.

from streetworks.streetmanager import StreetManagerClient, Environment

with StreetManagerClient("api-user@example.com", password, environment=Environment.SANDBOX) as sm:
    sm.authenticate()                                  # verify credentials
    submitted = sm.reporting.permits(status="submitted")

Shared across all modules: automatic retries with exponential backoff and jitter, Retry-After-aware 429 handling, a single exception hierarchy, and both sync and async clients built on httpx.

What this is (and isn't)

It is a typed client library: it handles authentication, token lifecycles, retries, rate limiting, pagination, and request/response plumbing for each of these APIs, so you call Python methods instead of hand-rolling HTTP. Auth and connectivity are verified against the real systems (see Status below).

It isn't a replacement for the APIs' own documentation. You still bring your own credentials (issued by the service operators, not by this SDK) and you still need each API's domain concepts — what a permit payload contains, what makes a valid USRN filter, which DataVIA layer holds which data. The SDK gets you connected and typed; the linked docs tell you what to send.

Status

Early alpha. Authentication and read/consume access are verified against the real systems for all providers: Street Manager (SANDBOX), Geoplace DataVIA (live — including a real feature query), D-TRO (production token + events search), the Open Data SNS parsing/verification pipeline, SRWR Open Data (parsed against real published daily and monthly extracts), OS Open USRN (Downloads API + GeoPackage reader), and UK Police (live safety_signal() and category queries against data.police.uk).

Not yet exercised against live systems — implemented to the published specs and covered by mocked tests: the write/publish paths (Street Manager work submission and assessment; D-TRO create/update and provisions). These are publisher-scoped and deliberately excluded from the read-only smoke test.

Known reconciliation items: D-TRO v4.0.0 schema models to follow when it lands (production cut-over expected mid-2026; v3.5.1 models ship now); the streetworks.exceptions API and client method surface may change before 1.0. See docs/INTEGRATION.md for how to verify against the real systems yourself. First-contact reports welcome.

Install

pip install streetworks            # core
pip install "streetworks[sns]"     # + SNS signature verification (cryptography)

Requires Python 3.10+.

Quickstart

The fastest way to see everything working: copy the credential template, fill in what you have, and run the one-file tour — it logs in to each configured provider and retrieves a little real data (read-only). Providers you leave blank are skipped, and SRWR / OS Open USRN need no credentials at all.

cp .env.example .env      # then edit .env
python examples/quickstart.py

For connectivity checks without data retrieval, use the smoke test instead: python scripts/smoke_test.py.

Prerequisites: credentials

Credentials are issued by the service operators. You only need the ones for the service(s) you'll use. Keep them in environment variables or a secret manager — never in code.

Credentials are per-environment — sandbox/integration credentials do not work against production, and vice versa.

Verify your setup

Before writing any code, confirm your credentials and connectivity with the included smoke test. It targets the test environments by default, is read-only, and skips any service you haven't configured:

SM_EMAIL='api-user@example.com' SM_PASSWORD='...' python scripts/smoke_test.py

================================================================
streetworks connectivity smoke test
TARGET  Street Manager: sandbox
All checks are READ-ONLY.
================================================================

  [PASS] Street Manager - authenticated (sandbox/v6), organisation 1355
  ...

A FAIL prints the exact exception, so a wrong credential or environment is obvious immediately. See docs/INTEGRATION.md for the full variable list and how to (deliberately) target production.

Street Manager

Authentication, token caching, and refresh (via the Party API, with automatic fall-back to re-authentication) are handled for you — following the DfT integration guidance: one token, reused, never re-authenticating per call.

from streetworks.streetmanager import StreetManagerClient, Environment, ApiVersion

with StreetManagerClient(
    "api-user@example.com",
    "password",                      # store securely, e.g. environment variable
    environment=Environment.SANDBOX, # or Environment.PRODUCTION
    version=ApiVersion.V6,           # or ApiVersion.V7 / ApiVersion.LATEST
) as sm:
    # Typed convenience methods for common workflows...
    work = sm.work.get_work("TSR1591199404915")
    submitted = sm.reporting.permits(status="submitted")

    # Or let the SDK walk every page for you:
    for permit in sm.reporting.iter_permits(status="submitted"):
        ...
    sm.work.assess_permit("TSR1591199404915", "TSR1591199404915-01",
                          {"assessment_status": "granted", ...})

    # ...and a generic escape hatch for every endpoint we haven't wrapped yet:
    s58 = sm.work.post("section-58s", json={...})
    updates = sm.event.works_updates()

Async is a mirror image:

from streetworks.streetmanager import AsyncStreetManagerClient

async with AsyncStreetManagerClient("api-user@example.com", "password") as sm:
    permits = await sm.reporting.permits(status="submitted")

Environments. Environment.SANDBOX and Environment.PRODUCTION are isolated systems with separate credentials. Develop and test against SANDBOX; only point at PRODUCTION once your workflows are proven. The smoke test and integration suite refuse to touch production without an explicit opt-in, so a stray setting can't send you at live data by accident.

Typed models

Pydantic v2 models generated from the official DfT swagger specifications live under streetworks.streetmanager.models.<version> and validate any client payload:

from streetworks.streetmanager.models.v6.work import WorkResponse

work = WorkResponse.model_validate(sm.work.get_work("TSR1591199404915"))

To regenerate after a DfT release, run the Regenerate Street Manager models workflow from the Actions tab (it opens a PR), or locally:

pip install -e ".[gen]"
python scripts/generate_models.py --version v6 --from-dir specs/streetmanager/v6

Street Manager Open Data (SNS push)

Open Data is a push model: Street Manager POSTs event notifications to an HTTPS endpoint you host. The receiver needs no credentials — messages are authenticated with AWS's public signing certificate (fetched over HTTPS), not a shared secret, so there's nothing to configure on the SDK side for parsing, verifying, or confirming. streetworks.opendata handles all of that, framework-agnostic:

from streetworks.opendata import handle

# inside your web handler, with the raw request body:
event = handle(request_body, expected_topic_arn="arn:aws:sns:eu-west-2:...:...")
if event is not None:               # None => subscription handshake, auto-confirmed
    print(event["event_type"], event["object_reference"])

See examples/opendata_fastapi.py for a complete FastAPI receiver.

Credentials nuance. Receiving Open Data needs no credentials. But note there are two distinct feeds: the fully public Open Data feed (this module), and a separate per-organisation API Notifications feed whose subscription is set up by calling an authenticated Street Manager endpoint (POST api-notifications/subscribe) — that setup step needs Street Manager credentials, though the messages, once flowing, are received the same credential-free way. This module handles the receiving side of both.

Geoplace DataVIA

Basic auth or OAuth2 client credentials (server-to-server), the full NSG layer catalogue (Layer.STREET_LINES, ESU_STREETS, ESU_ONE_WAY_EXEMPTIONS, and the Interest / Construction / Special Designation layers in all three geometry flavours), composable OGC filters, and transparent paging:

from streetworks.datavia import DataViaClient, Layer, filters

with DataViaClient(username="user", password="pass") as dv:      # or client_id=/client_secret=
    street = dv.street_by_usrn(4401245)
    nearby = dv.streets_near_point(-0.138405, 50.825181, 100)    # within 100m

    sed = dv.get_features(
        Layer.SPECIAL_DESIGNATION_LINES,
        filter_fragment=filters.and_(
            filters.intersects_polygon(ring),
            filters.property_equals("special_designation_code", 3),
        ),
    )

    for feature in dv.iter_features(Layer.ESU_STREETS, page_size=500):
        ...

POST GetFeature bodies match the shapes in the DataVIA documentation (WFS 1.1.0 + ogc:Filter); GET KVP with startIndex/count is also available via get_features_kvp(). Output formats: GeoJSON (default), OGRGML, SHAPEZIP, CSV, SPATIALITEZIP.

WMS (rendered map images)

The same endpoints also serve OGC WMS, so you can pull rendered map images of NSG layers or ask "what street is at this pixel?":

from pathlib import Path

png = dv.get_map([Layer.STREET_LINES], (424000, 533800, 426000, 535200))
Path("durham-streets.png").write_bytes(png)

info = dv.get_feature_info(Layer.STREET_LINES, (424000, 533800, 426000, 535200),
                           i=384, j=384)      # pixel coords in the image

Coordinates default to British National Grid (EPSG:27700), which sidesteps the WMS 1.3.0 lat/lon axis-order trap that bites with EPSG:4326.

DfT D-TRO

OAuth2 client credentials (30-minute tokens, cached and renewed automatically), x-app-id and per-request X-Correlation-ID headers handled for you:

from streetworks.dtro import DTROClient, Environment

with DTROClient(client_id, client_secret, app_id=app_id,
                environment=Environment.INTEGRATION) as dtro:
    events = dtro.search_events(since="2026-06-01T00:00:00", pageSize=50)
    record = dtro.get_dtro(events["events"][0]["id"])

    dtro.create_dtro(payload)                          # publisher scope
    dtro.create_dtro_from_file(big_json, gzip=True)    # large D-TROs
    signed = dtro.get_all_dtros_url()                  # full CSV extract

    dtro.schema_versions()                             # available schema versions
    dtro.search({...})                                 # search published D-TROs
    dtro.create_provisions([...], dtro_id="...")       # provisions (App-Id header handled)

Scottish Road Works Register (SRWR) Open Data

Scotland's national road works register publishes its full noticing data as daily Open Data extracts under the Open Government Licence v3 — no credentials required. streetworks.srwr downloads the archives and parses the multi-record-type CSV format (spec v2.02) into typed records, grouped into complete Activities:

from streetworks.srwr import SRWRClient, describe

with SRWRClient() as srwr:
    archive = srwr.download_daily("srwr-daily.zip")
    for activity in srwr.iter_activities(archive):
        phase = activity.phases[-1]
        print(activity.activity_id,
              describe("works_type", phase.works_type),
              describe("activity_status", phase.activity_status),
              phase.location)

Parsing streams (a 4-million-record monthly archive parses in well under a minute at ~30 MB memory). Monthly/yearly archives concatenate the daily extracts; latest_activities() applies the spec's most-recent-occurrence rule. Notices, phases, sites, inspections, FPNs, restrictions and reference data are all exposed; describe() translates the register's coded values.

The authenticated SRWR (Aurora) web-services API is available only to Scottish roads authorities and utilities and is not publicly documented, so it isn't covered. The Open Data feed carries the register's noticing data and needs no account.

OS Open USRN

Every Unique Street Reference Number in Great Britain, with street geometry, as Ordnance Survey OpenData — no credentials required. USRNs are the common key across this SDK: Street Manager works, DataVIA streets, D-TRO regulated places and SRWR activities all reference them. streetworks.openusrn downloads the GeoPackage via the OS Downloads API and queries it with the standard library only (no GDAL or geospatial stack):

from streetworks.openusrn import OpenUSRNClient, UsrnDatabase, extract_gpkg

with OpenUSRNClient() as client:
    archive = client.download("osopenusrn.zip")   # ~300 MB, streamed

with UsrnDatabase(extract_gpkg(archive)) as db:
    street = db.get(33909869)
    print(street.geometry)        # WKT, British National Grid (EPSG:27700)

DATEX II (European roadworks)

DATEX II is the European standard for traffic and roadworks data exchange, used by the National Access Points across Europe. streetworks.datex2 is a streaming, namespace-tolerant parser for SituationPublication roadworks — DATEX II v3 and v2 — plus source adapters. The first is the Netherlands' credential-free NDW open data (XML):

from streetworks.datex2 import NDWClient, iter_roadworks

with NDWClient() as ndw:
    feed = ndw.download_planned_works("ndw-planned.xml.gz")

for situation in iter_roadworks(feed):
    works = situation.roadworks[0]
    print(works.source_name, works.road_maintenance_type,
          works.validity.overall_start, works.location.point)

The parser streams (the ~170 MB Dutch national feed parses in seconds at ~35 MB memory) and normalises locations across referencing methods. Coordinates are WGS84 latitude/longitude — not the British National Grid used by the UK providers here.

National Highways (England's Strategic Road Network) publishes its DATEX II v3.4 extended profile as JSON, not XML, so it needs its own parsing path rather than the streaming XML parser above — streetworks.datex2.nationalhighways maps that JSON onto the same Situation/SituationRecord models. Needs a free subscription key from the developer portal; it pages through results automatically via the x-next cursor:

from streetworks.datex2 import ClosureType, NationalHighwaysClient

with NationalHighwaysClient(subscription_key) as nh:
    for situation in nh.iter_roadworks(ClosureType.PLANNED):
        works = situation.roadworks[0]
        print(works.cause_type, works.road_maintenance_type, works.location.point)

(Verified against the live API: it returns XML regardless of Accept headers unless you also send X-Response-MediaType: application/json — the client sends this for you.)

Northern Ireland & Wales (traveller-information RSS)

The remaining UK nations are covered by open RSS feeds — credential-free, but shallower data: these are traveller-information services (current and forthcoming closures as human-readable text), not works registers. Typed fields are best-effort extractions and the raw text is always preserved.

Northern Ireland — TrafficWatchNI (streetworks.trafficwatchni): DfI's Traffic Information & Control Centre feeds for roadworks, incidents and events; trunk roads and motorways NI-wide plus all roads in Greater Belfast, refreshed every 5 minutes. Attribution required: credit DfI TICC and preserve item URLs.

Wales — Traffic Wales (streetworks.trafficwales): Welsh Government feeds for roadworks, incidents/events and headlines on the motorway and trunk road network, in English and Welsh, refreshed every 5 minutes. Attribution required: credit Traffic Wales. (Traffic Wales also offers richer DATEX II feeds — access on application via traffic.wales/developers; once granted, streetworks.datex2 can parse them.)

from streetworks.trafficwatchni import TrafficWatchNIClient
from streetworks.trafficwales import TrafficWalesClient, Feed

with TrafficWatchNIClient() as twni:
    for item in twni.fetch():
        print(item.closure_type, item.road, item.town, "-", item.promoter)

with TrafficWalesClient() as tw:
    for item in tw.fetch(Feed.ROADWORKS):
        print(item.roads, item.title)

UK Police (crime data — a worker-safety signal)

There's no API for reporting abuse or aggression towards road workers directly — it doesn't exist. What does exist is the UK Police API (data.police.uk), which publishes street-level crime for England, Wales, and Northern Ireland — no credentials required. streetworks.police wraps it as a contextual safety signal for planning lone working or an unfamiliar site, not as a street-works dataset in its own right.

from streetworks.police import PoliceClient

with PoliceClient() as police:
    signal = police.safety_signal(51.500617, -0.124629)  # lat, lng of the worksite
    print(signal)
    # {'date': None, 'total_crimes': 3420, 'safety_relevant_count': 1623,
    #  'by_category': {'anti-social-behaviour': 1152, 'violent-crime': 344,
    #                  'public-order': 98, 'robbery': 21, 'possession-of-weapons': 8}}

safety_signal() fetches crime in roughly a one-mile radius of a point and counts only the categories in SAFETY_RELEVANT_CATEGORIES — violence and sexual offences, public order, anti-social behaviour, robbery, and possession of weapons. Property crime (vehicle crime, burglary, shoplifting, bicycle theft, criminal damage) is fetched but excluded from the count, because it says little about the risk of confrontation to a person on site. The raw per-point and per-polygon methods (street_level_crimes, street_level_crimes_in_area, crimes_at_location, crimes_no_location, forces, locate_neighbourhood, ...) are also available unfiltered.

Read this as contextual awareness, not prediction — three things that would otherwise mislead:

1. Historical, not live. The API publishes street-level crime roughly a month or two in arrears, aggregated per calendar month — recent past, not what's happening at the site today.
2. Area-level, not site-level. Police deliberately anonymise each crime's location to a snapped map point (often the middle of the street, sometimes 100m+ off the true spot) to protect victim privacy. This is a signal about the surrounding area, never the exact worksite.
3. Category matters more than the total. "High crime" as a lump figure is close to meaningless for personal safety — an area heavy in vehicle crime or shoplifting says little about risk to a road crew. That's why safety_signal() filters to the categories that actually bear on it rather than reporting the raw total.

Design principles

1. Never block the user. Typed methods for confirmed, common endpoints; generic get/post/put/delete on every API group for everything else.
2. Be a good API citizen. Token reuse, refresh-then-reauth, exponential backoff, honoured Retry-After — per the DfT integration guidance.
3. Test without credentials, verify with them. The whole unit suite runs against mocked transports (respx) so CI needs no secrets; a separate smoke test and skip-guarded integration suite verify against the real systems when you supply credentials.
4. Room to grow. Each provider is a self-contained module over a shared transport/exception core — adding a new API is additive.

Roadmap

• Pydantic model generation pipeline for the Street Manager swagger specs
• Auto-pagination helpers for the Reporting API (iter_permits() etc.)
• DataVIA WMS support (get_map, get_feature_info, wms_capabilities)
• D-TRO publish models generated from the DfT JSON schemas, version-namespaced (v3.5.1 to match production, v4.0.0 to follow) — see docs/DTRO_SCHEMAS.md
• Scottish Road Works Register - Open Data provider (streetworks.srwr). The authenticated SRWR/Aurora web-services API is restricted to Scottish authorities and utilities; contributions from SRWR users welcome.
• Common models: canonical cross-provider types (Street, WorksNotice, Coordinate, ...) with explicit .to_common() converters, so the same code handles English and Scottish data - native full-fidelity interfaces retained
• OS Open USRN: credential-free GB-wide USRN lookup with geometry (streetworks.openusrn)
• Northern Ireland roadworks (TrafficWatchNI RSS) and Wales motorway/trunk roadworks (Traffic Wales RSS) — all four UK nations now have coverage
• UK Police crime data (streetworks.police) as a worker-safety signal — no API exists for roadworker abuse directly, so this is the closest honest proxy; safety_signal() filters to the categories that bear on personal safety — verified against the real API
• Traffic Wales DATEX II feeds (richer than the RSS; access on application)
• Scottish street gazetteer (OSG portal open data); Northern Ireland gazetteer (Wales street gazetteer is already covered by the Geoplace NSG via DataVIA)
• DATEX II parser (v3 + v2 SituationPublication roadworks) with the NDW (Netherlands, XML) open-data adapter — verified against the real national feed
• National Highways (England SRN) DATEX II v3.4 JSON adapter (streetworks.datex2.nationalhighways), cursor pagination via x-next — verified against the real API
• Further DATEX II adapters: Mobilithek (DE), transport.data.gouv.fr (FR) — per-NAP verification needed
• Ordnance Survey NGD / Linked Identifiers?

0.4.0 — European & Crown Dependency roadworks

Candidate feeds, researched but not yet verified. As always, each needs a real sample feed and a licence/access check before building — the first task per source is "can we get the feed and what do the terms permit," not coding.

Grouped by the client shape they need:

• DATEX II adapters (thin fetchers over the existing streetworks.datex2 parser, NDW-style). Candidates: Finland (Digitraffic — open, well-documented, the natural next one to prove the pattern), Norway (Statens vegvesen), Denmark (Vejdirektoratet), Sweden (Trafikverket — verify its SOAP/XML model is DATEX-compatible), Spain (DGT NAP), France (Bison Futé). Access models vary from fully open to registration/agreement-gated — confirm per country.
• ArcGIS REST (a new client shape — Esri /query?f=json). Jersey publishes roadworks as an ArcGIS MapServer layer; likely a quick, self-contained win and the SDK's first Channel Islands coverage.
• Dedicated pieces (each its own project, not a quick adapter): Germany's Mobilithek (broker/subscription access, mixed schemas — D-TRO-scale effort); Guernsey (appears to be an HTML site — confirm whether any structured feed exists before committing, and check licensing for scraping).
• Verify-the-source-first: prefer official government feeds over third-party API-marketplace wrappers; a couple of the researched links need their real upstream endpoint confirmed.

US work zones (WZDx) — separate strand, own research session

The US standard for roadworks is WZDx (Work Zone Data Exchange), GeoJSON- based and distinct from DATEX II — so it needs its own parser, not a datex2 adapter. The USDOT WZDx feed registry is the canonical directory of live publishers and feed URLs — the right starting point for surveying what exists.

International gazetteers — separate strand

The European equivalents of OS Open USRN (address/street reference layers, not roadworks — keep distinct from the feeds above): France BAN, Spain Catastro, Norway Kartverket, Netherlands PDOK, Germany Geoportal, Portugal SNIG, plus the UK GeoPlace gazetteer SOAP API. These eventually connect to the common models work; formats differ widely, so each needs its own mapping design.

Contributions welcome — see CONTRIBUTING.md.

Development

pip install -e ".[dev]"
pytest                    # 35 mocked unit tests - no credentials needed
ruff check .

The unit tests mock the network so they run offline and without credentials. To verify the SDK against the real test/sandbox systems with your own credentials, use the smoke test or the integration suite — see docs/INTEGRATION.md:

python scripts/smoke_test.py     # one read-only call per configured service
pytest -m integration -v         # same checks, in the test suite

Licence

MIT. Not affiliated with or endorsed by the Department for Transport or Geoplace. Street Manager documentation is © Crown copyright, available under the Open Government Licence v3.0.