Fastapi

FastAPI (project skill)

Best-practice FastAPI patterns for this api-template. Pair with .claude/rules/backend/fastapi-routes.md (path-scoped route rules) and the pydantic-best-practices skill for models. The Cursor twin lives at .cursor/skills/fastapi/SKILL.md - keep the two in sync when updating either.

Essentials for this repo

• Validate path/query/body with Annotated[..., Path|Query|Body] or request models - never leave API inputs loosely typed.
• Return DTOs from app/dtos/; keep handlers thin; resolve config via Depends + get_settings from app/core/config.py.
• Raise CoreError subclasses (app/exceptions/) for domain failures; the global handler maps status codes.
• Inject the shared lifespan httpx2.AsyncClient via Depends(get_http_client) from app.core.http_client - never create a per-request client.
• Prefer Context7 / docs-researcher for version-sensitive FastAPI API questions over training memory.
• Finish with make check and the affected pytest modules.

Use the fastapi CLI

Run the development server on localhost with reload:

fastapi dev

Run the production server:

fastapi run

Not in this repo, though. fastapi-cli sits in the devserver dependency group here, never in [project] dependencies, so it is absent from the Docker runtime image: production runs uvicorn app.main:app directly, and make prod runs the same command locally (bound to loopback rather than 0.0.0.0 - see the comment on that target). Use fastapi dev / make dev for local development and leave fastapi run alone. See quality/python-tooling.

Add an entrypoint in pyproject.toml

FastAPI CLI will read the entrypoint in pyproject.toml to know where the FastAPI app is declared.

[tool.fastapi]
entrypoint = "my_app.main:app"

Use fastapi with a path

When adding the entrypoint to pyproject.toml is not possible, or the user explicitly asks not to, or it's running an independent small app, you can pass the app file path to the fastapi command:

fastapi dev my_app/main.py

Prefer to set the entrypoint in pyproject.toml when possible.

Use Annotated

Always prefer the Annotated style for parameter and dependency declarations.

It keeps the function signatures working in other contexts, respects the types, allows reusability.

In Parameter Declarations

Use Annotated for parameter declarations, including Path, Query, Header, etc.:

from typing import Annotated

from fastapi import FastAPI, Path, Query

app = FastAPI()


@app.get("/items/{item_id}")
async def read_item(
    item_id: Annotated[int, Path(ge=1, description="The item ID")],
    q: Annotated[str | None, Query(max_length=50)] = None,
):
    return {"message": "Hello World"}

instead of:

# DO NOT DO THIS
@app.get("/items/{item_id}")
async def read_item(
    item_id: int = Path(ge=1, description="The item ID"),
    q: str | None = Query(default=None, max_length=50),
):
    return {"message": "Hello World"}

For Dependencies

Use Annotated for dependencies with Depends().

Unless asked not to, create a new type alias for the dependency to allow re-using it.

from typing import Annotated

from fastapi import Depends, FastAPI

app = FastAPI()


def get_current_user():
    return {"username": "johndoe"}


CurrentUserDep = Annotated[dict, Depends(get_current_user)]


@app.get("/items/")
async def read_item(current_user: CurrentUserDep):
    return {"message": "Hello World"}

instead of:

# DO NOT DO THIS
@app.get("/items/")
async def read_item(current_user: dict = Depends(get_current_user)):
    return {"message": "Hello World"}

Do not use Ellipsis for path operations or Pydantic models

Do not use ... as a default value for required parameters, it's not needed and not recommended.

Do this, without Ellipsis (...):

from typing import Annotated

from fastapi import FastAPI, Query
from pydantic import BaseModel, Field


class Item(BaseModel):
    name: str
    description: str | None = None
    price: float = Field(gt=0)


app = FastAPI()


@app.post("/items/")
async def create_item(item: Item, project_id: Annotated[int, Query()]): ...

instead of this:

# DO NOT DO THIS
class Item(BaseModel):
    name: str = ...
    description: str | None = None
    price: float = Field(..., gt=0)


app = FastAPI()


@app.post("/items/")
async def create_item(item: Item, project_id: Annotated[int, Query(...)]): ...

Return Type or Response Model

When possible, include a return type. It will be used to validate, filter, document, and serialize the response.

from fastapi import FastAPI
from pydantic import BaseModel

app = FastAPI()


class Item(BaseModel):
    name: str
    description: str | None = None


@app.get("/items/me")
async def get_item() -> Item:
    return Item(name="Plumbus", description="All-purpose home device")

Important: Return types or response models are what filter data ensuring no sensitive information is exposed. And they are used to serialize data with Pydantic (in Rust), this is the main idea that can increase response performance.

The return type doesn't have to be a Pydantic model, it could be a different type, like a list of integers, or a dict, etc.

When to use response_model instead

If the return type is not the same as the type that you want to use to validate, filter, or serialize, use the response_model parameter on the decorator instead.

from typing import Any

from fastapi import FastAPI
from pydantic import BaseModel

app = FastAPI()


class Item(BaseModel):
    name: str
    description: str | None = None


@app.get("/items/me", response_model=Item)
async def get_item() -> Any:
    return {"name": "Foo", "description": "A very nice Item"}

This can be particularly useful when filtering data to expose only the public fields and avoid exposing sensitive information.

from typing import Any

from fastapi import FastAPI
from pydantic import BaseModel

app = FastAPI()


class InternalItem(BaseModel):
    name: str
    description: str | None = None
    secret_key: str


class Item(BaseModel):
    name: str
    description: str | None = None


@app.get("/items/me", response_model=Item)
async def get_item() -> Any:
    item = InternalItem(
        name="Foo", description="A very nice Item", secret_key="supersecret"
    )
    return item

Performance

Do not use ORJSONResponse or UJSONResponse, they are deprecated.

Instead, declare a return type or response model. Pydantic will handle the data serialization on the Rust side.

Including Routers

When declaring routers, prefer to add router level parameters like prefix, tags, etc. to the router itself, instead of in include_router().

Do this:

from fastapi import APIRouter, FastAPI

app = FastAPI()

router = APIRouter(prefix="/items", tags=["items"])


@router.get("/")
async def list_items():
    return []


# In main.py
app.include_router(router)

instead of this:

# DO NOT DO THIS
from fastapi import APIRouter, FastAPI

app = FastAPI()

router = APIRouter()


@router.get("/")
async def list_items():
    return []


# In main.py
app.include_router(router, prefix="/items", tags=["items"])

There could be exceptions, but try to follow this convention.

Apply shared dependencies at the router level via dependencies=[Depends(...)].

Dependency Injection

See the dependency injection reference for detailed patterns including yield with scope, and class dependencies.

Use dependencies when the logic can't be declared in Pydantic validation, depends on external resources, needs cleanup (with yield), or is shared across endpoints.

Apply shared dependencies at the router level via dependencies=[Depends(...)].

Async vs Sync path operations

Use async path operations only when fully certain that the logic called inside is compatible with async and await (it's called with await) or that doesn't block.

from fastapi import FastAPI

app = FastAPI()


# Use async def when calling async code
@app.get("/async-items/")
async def read_async_items():
    data = await some_async_library.fetch_items()
    return data


# Use plain def when calling blocking/sync code or when in doubt
@app.get("/items/")
def read_items():
    data = some_blocking_library.fetch_items()
    return data

In case of doubt, or by default, use regular def functions, those will be run in a threadpool so they don't block the event loop.

The same rules apply to dependencies.

Make sure blocking code is not run inside of async functions. The logic will work, but will damage the performance heavily.

When needing to mix blocking and async code, see Asyncer in the other tools reference.

Streaming (JSON Lines, SSE, bytes)

See the streaming reference for JSON Lines, Server-Sent Events (EventSourceResponse, ServerSentEvent), and byte streaming (StreamingResponse) patterns.

Tooling

See the other tools reference for details on uv, Ruff, ty for package management, linting, type checking, formatting, etc.

Other Libraries

See the other tools reference for details on other libraries:

• Asyncer for handling async and await, concurrency, mixing async and blocking code, prefer it over AnyIO or asyncio.
• SQLModel for working with SQL databases, prefer it over SQLAlchemy.
• HTTPX2 for interacting with HTTP (other APIs), prefer it over Requests and classic httpx.

Do not use Pydantic RootModels

Do not use Pydantic RootModel, instead use regular type annotations with Annotated and Pydantic validation utilities.

For example, for a list with validations you could do:

from typing import Annotated

from fastapi import Body, FastAPI
from pydantic import Field

app = FastAPI()


@app.post("/items/")
async def create_items(items: Annotated[list[int], Field(min_length=1), Body()]):
    return items

instead of:

# DO NOT DO THIS
from typing import Annotated

from fastapi import FastAPI
from pydantic import Field, RootModel

app = FastAPI()


class ItemList(RootModel[Annotated[list[int], Field(min_length=1)]]):
    pass


@app.post("/items/")
async def create_items(items: ItemList):
    return items

FastAPI supports these type annotations and will create a Pydantic TypeAdapter for them, so that types can work as normally and there's no need for the custom logic and types in RootModels.

Use one HTTP operation per function

Don't mix HTTP operations in a single function, having one function per HTTP operation helps separate concerns and organize the code.

Do this:

from fastapi import FastAPI
from pydantic import BaseModel

app = FastAPI()


class Item(BaseModel):
    name: str


@app.get("/items/")
async def list_items():
    return []


@app.post("/items/")
async def create_item(item: Item):
    return item

instead of this:

# DO NOT DO THIS
from fastapi import FastAPI, Request
from pydantic import BaseModel

app = FastAPI()


class Item(BaseModel):
    name: str


@app.api_route("/items/", methods=["GET", "POST"])
async def handle_items(request: Request):
    if request.method == "GET":
        return []