Skip to content

Latest commit

 

History

History
197 lines (138 loc) · 5.75 KB

File metadata and controls

197 lines (138 loc) · 5.75 KB

openapi-python

openapi-python generates typed Python API clients from OpenAPI specs, with a developer-friendly and ergonomic string-literal-based interface strongly inspired by openapi-typescript.

openapi-python demo

Installation

# If you want to define your own HTTP transport (requests, asyncio, ...)
uv add openapi-python

# For built-in httpx transport:
uv add openapi-python[httpx]

Client generation

Generate a client from an OpenAPI spec in openapi.json:

# Types + HTTP client/transport
uv run openapi-python generate --spec ./openapi.json --out ./generated

# Just types, use your own HTTP client/transport
uv run openapi-python generate --spec ./openapi.json --out ./generated --transport-mode protocol-only

... or programatically:

from pathlib import Path
from openapi_python import GenerationRequest, generate_client

result = generate_client(
    GenerationRequest(
        spec_source="./openapi.json",
        output_dir=Path("./generated"),
        package_name="my_client",
        overwrite=True,
    )
)

Using generated clients

Generated clients expose route-specific callables with typed params, query, headers, body, and return values.

With the built-in httpx transport:

from generated.my_client import Client

client = Client(base_url="https://api.example.com")
book = client.get("/books/{book_id}")(params={"book_id": 1})

For async APIs:

from generated.my_client import AsyncClient

async_client = AsyncClient(base_url="https://api.example.com")
book = await async_client.get("/books/{book_id}")(params={"book_id": 1})

For protocol-only clients, provide your own transport:

from generated.my_client import Client

client = Client(base_url="https://api.example.com", transport=my_transport)
book = client.get("/books/{book_id}")(params={"book_id": 1})

Extensibility

GeneratorExtensions exposes two safe hooks:

  • normalize_hooks: transform the normalized model before rendering.
  • render_context_hooks: transform rendered file content map before writing.

Transport Decoupling

Generated clients expose a transport protocol. You can plug in your own transport while keeping route-level typing guarantees.

Use --transport-mode protocol-only to generate clients that require a supplied transport and do not emit the built-in httpx transport classes. The default --transport-mode default includes DefaultTransport and DefaultAsyncTransport, which require the httpx extra when instantiated.

Built-in httpx transport

Install the httpx extra and generate with the default transport mode:

uv add "openapi-python[httpx]"
uv run openapi-python generate --spec ./openapi.json --out ./generated --package my_client

You can supply preconfigured httpx clients:

import httpx

from generated.my_client import AsyncClient, Client, DefaultAsyncTransport, DefaultTransport

sync_http = httpx.Client(headers={"authorization": "Bearer token"})
async_http = httpx.AsyncClient(headers={"authorization": "Bearer token"})

client = Client(
    base_url="https://api.example.com",
    transport=DefaultTransport(client=sync_http),
)
async_client = AsyncClient(
    base_url="https://api.example.com",
    transport=DefaultAsyncTransport(client=async_http),
)

Custom transport

Install openapi-python without extras and generate protocol-only code:

uv add openapi-python requests
uv run openapi-python generate \
  --spec ./openapi.json \
  --out ./generated \
  --package my_client \
  --transport-mode protocol-only

Then provide an object that satisfies the generated Transport protocol:

from collections.abc import Mapping

import requests

from generated.my_client import Client


class RequestsTransport:
    def request(
        self,
        *,
        method: str,
        route: str,
        base_url: str,
        params: Mapping[str, object] | None,
        query: Mapping[str, object] | None,
        headers: Mapping[str, object] | None,
        body: object | None,
    ) -> object:
        response = requests.request(
            method=method.upper(),
            url=f"{base_url.rstrip('/')}{route.format(**(params or {}))}",
            params={key: str(value) for key, value in (query or {}).items()} or None,
            headers={key: str(value) for key, value in (headers or {}).items()} or None,
            json=body,
        )
        response.raise_for_status()
        if response.content:
            return response.json()
        return None


client = Client(
    base_url="https://api.example.com",
    transport=RequestsTransport(),
)
book = client.get("/books/{book_id}")(params={"book_id": 1})

Releases

Releases are published from the protected releases branch. The package version is set manually in pyproject.toml, and pushing a release commit to releases triggers the GitHub Actions release workflow. The workflow creates the matching vX.Y.Z tag after checks pass.

Before the first release, configure PyPI Trusted Publishing for this repository:

  • PyPI project: openapi-python
  • GitHub workflow: release.yml
  • GitHub environment: pypi

The GitHub pypi environment should be limited to deployments from the releases branch.

Release steps:

# 1. Update project.version in pyproject.toml, then commit that change.
uv run python scripts/release.py --version 0.1.0

# 2. If checks pass, push the current commit to the releases branch.
uv run python scripts/release.py --version 0.1.0 --push-release-branch

The release workflow verifies that the version tag does not already exist, runs checks, builds the distributions, validates them with twine, creates the release tag, publishes to PyPI, and creates a GitHub Release with generated notes.