Generate modern Python clients from OpenAPI 3.0 and 3.1 documents.
This generator does not support OpenAPI 2.x FKA Swagger. If you need to use an older document, try upgrading it to version 3 first with one of many available converters.
This project is still in development and does not support all OpenAPI features
This tool focuses on creating the best developer experience for Python developers by:
- Using all the latest and greatest Python features like type annotations and dataclasses.
- Having documentation and usage instructions specific to this one generator.
- Being written in Python with Jinja2 templates, making it easier to improve and extend for Python developers. It's also much easier to install and use if you already have Python.
I recommend you install with pipx so you don't conflict with any other packages you might have: pipx install openapi-python-client --include-deps.
Note the
--include-depsoption makesruffavailable in your path so thatopenapi-python-clientcan use it to clean up the generated code.
If you use pipx run then the post-generation hooks will not be available unless you install them manually.
You can also install with normal pip: pip install openapi-python-client
Then, if you want tab completion: openapi-python-client --install-completion
openapi-python-client generate --url https://my.api.com/openapi.json
This will generate a new client library named based on the title in your OpenAPI spec. For example, if the title of your API is "My API", the expected output will be "my-api-client". If a folder already exists by that name, you'll get an error.
If you have an openapi.json file available on disk, in any CLI invocation you can build off that instead by replacing --url with a --path:
openapi-python-client generate --path location/on/disk/openapi.json
openapi-python-client update --url https://my.api.com/openapi.json
This feature leverages Jinja2's ChoiceLoader and FileSystemLoader. This means you do not need to customize every template. Simply copy the template(s) you want to customize from the default template directory to your own custom template directory (file names must match exactly) and pass the template directory through the custom-template-path flag to the generate and update commands. For instance,
openapi-python-client update \
--url https://my.api.com/openapi.json \
--custom-template-path=relative/path/to/mytemplates
Be forewarned, this is a beta-level feature in the sense that the API exposed in the templates is undocumented and unstable.
- A
pyproject.tomlfile, optionally with Poetry metadata (default), PDM (with--meta=pdm), or only Ruff config. - A
README.mdyou'll most definitely need to update with your project's details - A Python module named just like the auto-generated project name (e.g. "my_api_client") which contains:
- A
clientmodule which will have both aClientclass and anAuthenticatedClientclass. You'll need these for calling the functions in theapimodule. - An
apimodule which will contain one module for each tag in your OpenAPI spec, as well as adefaultmodule for endpoints without a tag. Each of these modules in turn contains one function for calling each endpoint. - A
modelsmodule which has all the classes defined by the various schemas in your OpenAPI spec
- A
- A
setup.pyfile if you use--meta=setup(default is--meta=poetry)
For a full example you can look at the end_to_end_tests directory which has baseline_openapi_3.0.json and baseline_openapi_3.1.yaml files.
The "golden-record" in that same directory is the generated client from either of those OpenAPI documents.
You can pass a YAML (or JSON) file to openapi-python-client with the --config option in order to change some behavior.
The following parameters are supported:
Used to change the name of generated model classes. This param should be a mapping of existing class name (usually a key in the "schemas" section of your OpenAPI document) to class_name and module_name. As an example, if the name of a model in OpenAPI (and therefore the generated class name) was something like "_PrivateInternalLongName" and you want the generated client's model to be called "ShortName" in a module called "short_name" you could do this:
Example:
class_overrides:
_PrivateInternalLongName:
class_name: ShortName
module_name: short_nameThe easiest way to find what needs to be overridden is probably to generate your client and go look at everything in the models folder.
Used to change the name of generated client library project/package. If the project name is changed but an override for the package name
isn't provided, the package name will be converted from the project name using the standard convention (replacing -'s with _'s).
Example:
project_name_override: my-special-project-name
package_name_override: my_extra_special_package_nameWhen generating properties, the name attribute of the OpenAPI schema will be used. When the name is not a valid Python identifier (e.g. begins with a number) this string will be prepended. Defaults to "field_". It will also be used to prefix fields in schema starting with "_" in order to avoid ambiguous semantics.
Example:
field_prefix: attr_Specify the package version of the generated client. If unset, the client will use the version of the OpenAPI spec.
Example:
package_version_override: 1.2.3In the config file, there's an easy way to tell openapi-python-client to run additional commands after generation. Here's an example showing the default commands (using Ruff) that will run if you don't override them in config:
post_hooks:
- "ruff check . --fix"
- "ruff format ."By default, openapi-python-client generates class names which include the full path to the schema, including any parent-types. This can result in very long class names like MyRouteSomeClassAnotherClassResponse—which is very unique and unlikely to cause conflicts with future API additions, but also super verbose.
If you are carefully curating your title properties already to ensure no duplicate class names, you can turn off this prefixing feature by setting use_path_prefixes_for_title_model_names to false in your config file. This will use the title property of any object that has it set without prefixing.
If this option results in conflicts, you will need to manually override class names instead via the class_overrides option.
By default, the timeout for retrieving the schema file via HTTP is 5 seconds. In case there is an error when retrieving the schema, you might try and increase this setting to a higher value.