self-documenting-apis

Self-Documenting APIs

Overview

FastAPI auto-generates OpenAPI docs at /docs (Swagger UI) and /redoc (ReDoc). When endpoints have proper docstrings and typed models, these docs are comprehensive enough to replace any hand-written API reference. The goal: never maintain a separate api-reference.md — the code IS the documentation.

When to Use

Apply this skill when:

• Creating a new API endpoint
• Refactoring an existing endpoint (changing request/response shape)
• Adding a new router module
• Any milestone in a plan that involves API changes

Skip when:

• Internal helper functions (not exposed via HTTP)
• WebSocket endpoints (OpenAPI support is limited)

The Practices

1. Every Endpoint Gets a Docstring

The docstring becomes the endpoint description in Swagger UI. Write it for the developer consuming your API.

@router.post("/v1/connections/oauth")
async def create_oauth_session(
    body: OAuthSessionRequest,
    developer: CurrentDeveloper,
):
    """Start OAuth flow for an end user.

    Creates a Nango Connect Session and returns a URL to redirect the end user to.
    The end user completes OAuth in their browser, then call POST /v1/connections/confirm.
    """

Rules:

• First line: what the endpoint does (imperative mood)
• Additional lines: context a consumer needs — what happens next, side effects, prerequisites
• Don't document request/response fields in the docstring — that's what models are for

2. Typed Response Models (Not Raw Dicts)

Never return raw dicts. Define a Pydantic model so the response schema appears in docs.

# Bad — docs show no response schema
@router.get("/v1/connections/{id}")
async def get_connection(...):
    return {"id": conn.id, "status": conn.status}

# Good — docs show full response schema with field descriptions
class ConnectionResponse(BaseModel):
    id: str
    end_user_id: str
    provider: str
    status: str = Field(description="Connection status: active, error, deleting")
    created_at: datetime

@router.get("/v1/connections/{id}", response_model=ConnectionResponse)
async def get_connection(...) -> ConnectionResponse:
    ...

Key points:

• Set response_model= on the decorator so FastAPI validates AND documents the response
• Use Field(description=...) for non-obvious fields
• For lists: response_model=list[ConnectionResponse]

3. Typed Request Models with Field Descriptions

class SyncRequest(BaseModel):
    connection_id: str = Field(description="UUID of the connection to sync")
    resources: list[str] = Field(
        description="Resources to sync (e.g., ['tickets', 'articles'])"
    )

Use Field for:

• Non-obvious field names
• Fields with constraints (ge=1, le=1000)
• Fields with default values that need explanation
• Enum-like string fields (list the valid values)

4. Status Codes and Error Responses

Declare non-200 status codes so they appear in docs:

@router.post(
    "/v1/syncs/{id}/trigger",
    status_code=202,
    responses={409: {"description": "Sync already running"}},
)
async def trigger_sync(...):
    """Trigger a sync run in the background."""

5. Router Tags for Grouping

Group related endpoints with tags so Swagger UI organizes them into sections:

router = APIRouter(tags=["Connections"])

This replaces the need for section headers in a manual API reference.

6. Paginated List Responses

For list endpoints, use a consistent wrapper:

class PaginatedResponse(BaseModel, Generic[T]):
    items: list[T]
    total: int
    limit: int
    offset: int

This documents the pagination contract once, consistently.

Verification

After creating or modifying endpoints, check the auto-docs:

1. Start the server
2. Open /docs (Swagger UI) or /redoc
3. Verify:
   • Endpoint has a clear description (from docstring)
   • Request body schema shows all fields with descriptions
   • Response schema shows all fields with types
   • Status codes are documented
   • Endpoints are grouped logically (tags)

If a consumer can't understand the endpoint from /docs alone, the docstring or model is missing something.

Quick Reference

1. Docstring   — what it does, context, side effects
2. Response    — Pydantic model with Field(description=...)
3. Request     — Pydantic model with Field(description=...)
4. Status      — response_model=, status_code=, responses={}
5. Tags        — APIRouter(tags=["Section"])
6. Verify      — check /docs after changes

Anti-Patterns

Don't	Do
Return raw dicts	Define a response model
Write a separate api-reference.md	Let FastAPI generate docs from code
Document fields in the docstring	Use Field(description=...) on the model
Skip the docstring ("it's obvious")	Write it — /docs is your public contract
Use Any or dict as response type	Define a typed model, even for simple responses
Maintain two sources of truth	Code is the docs — update the code, docs update automatically