Python API reference
Constructors, lifecycle, catalog registry, discovery, search, and result methods.
Import the public types with from superstac import Client, AsyncClient. Both clients accept storage="memory"; other storage backends are rejected by the current Python bindings.
Constructors
Client(config=None, *, catalogs=None, providers=None, settings=None, storage="memory")
Client.open(url, *, id=None, storage="memory")
Client.from_yaml(yaml_path, *, storage="memory")AsyncClient has the same signatures. Await only AsyncClient.open(...); its constructor and from_yaml() are synchronous.
config is a dictionary with optional catalogs, providers, and settings entries. Supplying it takes precedence over the separate constructor keywords. Without an explicit ID, open() derives one from the first dot-separated hostname label.
Lifecycle and settings
| Method | Behavior |
|---|---|
start() | Initial health checks and collection discovery; no-op while started. |
shutdown() | Cancel monitors and mark the engine unstarted. |
get_settings() | Return the stored settings dictionary. |
update_settings(update) | Apply supported non-null fields from a dictionary. |
On AsyncClient, await start() and shutdown(). Read health limitations before relying on global monitoring settings.
Catalog registry
| Method | Input / output |
|---|---|
add_catalog(catalog, *, provider=None) | Add one catalog dictionary; return created catalog. |
add_catalogs(catalogs) | Add a list of dictionaries, honoring each provider field. |
get_catalog(id) | Return one catalog dictionary. |
list_catalogs() | Return registered catalogs. |
update_catalog(id, update) | Update supported catalog fields. |
delete_catalog(id) | Remove a catalog. |
Always include id and url when creating a catalog. update_catalog accepts provider, title, description, url, and settings. Alias maps are configured at creation, not through this update shape. Omitted title/description can be cleared by the current implementation.
Provider registry
add_provider(provider), add_providers(providers), get_provider(id), list_providers(), update_provider(id, update), and delete_provider(id) operate on descriptive provider records. Register a provider before linking catalogs to it.
These registry methods remain synchronous on AsyncClient.
Discovery
| Method | Result |
|---|---|
list_collections() | List of {"id": collection_id, "catalogs": [catalog_id, ...]}. |
collections_by_catalog() | Dictionary of catalog IDs to collection ID lists. |
catalogs_supporting(collection_id) | Catalog IDs advertising the collection. |
describe_collection(catalog_id, collection_id) | Full collection dictionary or None for a 404. |
get_collection(id) | First matching collection document; raises KeyError if none is found. |
get_collections() | Full documents for discovered collections; potentially many requests. |
Await all discovery methods on AsyncClient.
Search
client.search(
collections=["sentinel-2-l2a"],
bbox=[6.0, 49.0, 7.0, 50.0],
datetime="2024-01-01T00:00:00Z/2024-01-31T23:59:59Z",
limit=10,
)Pass collections explicitly, using [] for all collections. See search parameters for ids, intersects, limit semantics, and unsupported parameters. The return value is a Search object; await the call on AsyncClient.
Search results
items(), matched(), to_geojson(), item_collection_as_dict(), len(search), and the metadata property are synchronous. See result fields.
Python item dictionaries do not expose the Rust SearchItem.catalog_id / seen_in wrapper. Metadata includes run-level failures and counts, not per-item provenance.
Errors
Invalid Python input generally raises ValueError. Engine errors are mapped to RuntimeError; get_collection() uses KeyError for a missing collection. Per-catalog search failures can instead be returned in search.metadata["failures"], so catching exceptions alone does not establish completeness.
