UMLS Python Client

Production-ready Python SDK for the National Library of Medicine UMLS Terminology Services APIs.

umls-python-client keeps the original UMLSClient interface for existing users and adds typed sync, async, export, release, and workflow helpers for new applications. It is designed for developers and researchers who need a practical one-stop Python client for UMLS search, concepts, source vocabularies, semantic types, crosswalks, metadata, and UTS release downloads.

Why This Client

• Current NLM REST coverage, including the May 4, 2026 search filters semanticTypes and semanticGroups.
• Typed sync and async clients built on httpx.
• Backward-compatible legacy client with the original camelCase namespaces.
• Structured exceptions for modern clients and legacy error payloads for older code.
• JSON, JSONL, CSV, and RDF/Turtle exports.
• Practical helpers for bulk search, bulk crosswalk, concept profiles, source lookup, UMLS URL following, API-key validation, and release downloads.
• CI across Python 3.9 through 3.14.

Installation

pip install umls-python-client

For local development:

pip install ".[dev,docs]"

Authentication

Most UMLS requests require an API key from your UTS profile: https://documentation.uts.nlm.nih.gov/rest/authentication.html

import os

api_key = os.environ["UMLS_API_KEY"]

The metadata sources endpoint is public and does not require authentication, but the client still uses one consistent client configuration.

Typed Sync Quickstart

from umls_python_client import TypedUMLSClient

with TypedUMLSClient(api_key="YOUR_UMLS_API_KEY") as client:
    response = client.search_api.search(
        "diabetes",
        semantic_groups="Disorders",
        page_size=25,
    )

    for result in response.result:
        print(result.ui, result.name)

Typed methods return UMLSResponse[T]. Use response.result for parsed models, response.raw or response.to_dict() for the original UMLS envelope, and response.to_rdf() for Turtle RDF.

Async Quickstart

from umls_python_client import AsyncUMLSClient


async def main() -> None:
    async with AsyncUMLSClient(api_key="YOUR_UMLS_API_KEY") as client:
        response = await client.crosswalk_api.get_crosswalk(
            source="HPO",
            id="HP:0001947",
            target_source="SNOMEDCT_US",
        )
        print(response.result)

Legacy Compatible Usage

UMLSClient keeps existing imports, JSON string defaults, and camelCase namespaces:

from umls_python_client import UMLSClient

client = UMLSClient(api_key="YOUR_UMLS_API_KEY")

pretty_json = client.searchAPI.search("diabetes")
raw_dict = client.search_api.search("diabetes", return_indented=False)
rdf = client.cui_api.get_cui_info("C0011849", format="rdf")

Existing namespaces are preserved:

• client.searchAPI
• client.sourceAPI
• client.cuiAPI
• client.semanticNetworkAPI
• client.crosswalkAPI
• client.metadataAPI
• client.atomAPI
• client.authAPI
• client.releaseAPI

Modern namespaces are also available:

• search_api
• source_api
• cui_api
• semantic_network_api
• crosswalk_api
• metadata_api
• atom_api
• auth_api
• release_api

Common Recipes

Semantic Search

response = client.search_api.search(
    "depression",
    semantic_types="T047|T046",
    semantic_groups="Disorders",
)

semantic_types supports semantic type TUIs, semantic type names, and semantic tree number prefixes. Use pipes for multiple values, matching NLM's search documentation.

Source Code To CUI

response = client.search_api.search(
    "9468002",
    input_type="sourceUi",
    search_type="exact",
    sabs="SNOMEDCT_US",
)

HPO To SNOMED CT Crosswalk

response = client.crosswalk_api.get_crosswalk(
    source="HPO",
    id="HP:0001947",
    target_source="SNOMEDCT_US",
)

UMLS CUI crosswalks are useful starting points for mapping work. Review results before using them in clinical or research pipelines.

Concept Profile

profile = client.cui_api.get_concept_profile(
    "C0011849",
    include_atoms=False,
).result

print(profile.concept.name)
print([definition.value for definition in profile.definitions])

Metadata Lookup

sources = client.metadata_api.find_source(abbreviation="SNOMEDCT_US").result

Bulk Workflows

searches = client.search_api.bulk_search(["diabetes", "asthma"])
crosswalks = client.crosswalk_api.bulk_crosswalk(
    ["HP:0001947", "HP:0001250"],
    source="HPO",
    target_source="SNOMEDCT_US",
)

Follow UMLS URLs

Many UMLS payload fields contain URLs to related resources. Follow those URLs with the same authenticated client:

concept = client.cui_api.get_cui_info("C0011849").result
atoms = client.follow_url(concept.raw["atoms"])

Release Downloads

releases = client.release_api.list_releases(
    release_type="umls-full-release",
    current=True,
)

client.release_api.download_file(
    url=releases.result[0].url,
    path="downloads",
)

Call list_releases() without a release_type to inspect NLM's release catalog, or pass a release type such as umls-full-release to discover current download URLs.

Exports

Typed responses can save themselves:

response = client.search_api.search("fracture", page_size=25)

response.save("exports/search", format="jsonl")
response.save("exports/search.csv", format="csv")
response.save("exports/search.ttl", format="rdf")

Clients can export any response or raw payload:

client.export(response, "exports/search.json", format="json", overwrite=True)

Supported formats:

If the path has a file suffix, the client writes exactly there. If the path is a directory or suffixless path, parent directories are created and a safe filename is generated.

Legacy save_to_file=True still works and now delegates to the same export implementation.

Endpoint Coverage

Error Handling

Typed clients raise structured exceptions:

• UMLSRequestError
• UMLSHTTPError
• UMLSDecodeError

Legacy UMLSClient methods keep returning error payload dictionaries instead of raising for UMLS request failures.

Configure timeouts and retries at client construction:

client = TypedUMLSClient(
    api_key="YOUR_UMLS_API_KEY",
    timeout=20.0,
    retries=3,
)

Testing

ruff check .
ruff format --check .
mypy src
pytest
mkdocs build --strict
python -m build

Live integration tests are optional and require UMLS_API_KEY. Tests that would validate user keys or download release files remain opt-in and do not run in default CI.

Migration Notes

From 1.0.x and 1.1.0:

• Existing imports and UMLSClient behavior remain supported.
• Prefer TypedUMLSClient for new sync code.
• Prefer AsyncUMLSClient for async applications.
• Prefer response.save(...) or client.export(...) over new uses of save_to_file=True.
• Use snake_case namespaces in new code. CamelCase namespaces remain available for compatibility.

Documentation

• NLM UMLS REST docs: https://documentation.uts.nlm.nih.gov/rest/home.html
• NLM auth docs: https://documentation.uts.nlm.nih.gov/rest/authentication.html
• Project source: https://github.com/palasht75/umls-python-client

Releases

This repository uses Release Please and PyPI trusted publishing. Merged commits should use Conventional Commit prefixes:

• fix: for patch releases
• feat: for minor releases
• feat!: or fix!: for major releases

Release Please opens a release PR. Merging that release PR creates the GitHub Release, builds distributions, attaches artifacts, and publishes to PyPI.

GitHub Actions must be allowed to create release PRs, or the repository must provide a RELEASE_PLEASE_TOKEN secret with PR permissions.

License

Apache-2.0. See LICENSE.