growth-loop-sdk 0.1.4

Growth Python SDK — drop-in product analytics for vibe-coded SaaS. Pairs with the Growth MCP server to give Claude Code an AI growth engineer for your Python app.

growth-loop-sdk

Python SDK for growth-loop.dev — Claude Code-native product analytics for vibe-coded SaaS. The Python module name is growth; the PyPI distribution is growth-loop-sdk.

Pairs with @growth-loop/mcp-server so Claude Code can ask your funnels, retention, and revenue questions directly with citations.

Install

pip install growth-loop-sdk
# or
uv add growth-loop-sdk

Requires Python 3.10+.

Quick start

import os
from growth import Growth, GrowthOptions

growth = Growth(GrowthOptions(
    api_key=os.environ["GROWTH_KEY"],
    host=os.environ.get("GROWTH_HOST", "https://api.growth-loop.dev"),
    environment=os.environ.get("APP_ENV", "production"),
    release=os.environ.get("GIT_COMMIT_SHA"),
))

# Anywhere in your app:
growth.track("signup_completed", {"plan": "pro"})

The SDK runs a background worker thread; on atexit it flushes pending events automatically. On serverless platforms (Lambda, Cloud Run) call growth.flush() before the request handler returns.

FastAPI

from fastapi import APIRouter
from growth import growth_span, growth_step
from growth_setup import growth      # the singleton from above

router = APIRouter()

@router.post("/checkout")
@growth_span(growth, "checkout", distinct_id=lambda body: body.user_id)
async def checkout(body: CheckoutBody):
    # automatically emits checkout.started / .completed / .failed
    # with duration_ms and error metadata
    return await stripe.charge(body)

@router.post("/signup")
@growth_step(growth, "signup_completed", distinct_id=lambda body: body.user_id)
async def signup(body: SignupBody):
    return await create_user(body)

Flask

from flask import Blueprint
from growth import growth_span
from growth_setup import growth

bp = Blueprint("checkout", __name__)

@bp.route("/checkout", methods=["POST"])
@growth_span(growth, "checkout", distinct_id=lambda: g.user.id)
def checkout():
    return stripe.charge(request.json)

Core API

Growth(options: GrowthOptions)

@dataclass
class GrowthOptions:
    api_key: str                       # required
    host: str = "https://api.growth-loop.dev"
    flush_interval: float = 5.0        # seconds
    batch_size: int = 50
    environment: str | None = None     # "production" | "preview" | "development"
    release: str | None = None         # git SHA — used by /diagnose for commit-level attribution
    timeout: float = 5.0               # HTTP timeout

Method	What it does
track(name, properties=None, options=None)	Enqueue one event. Non-blocking.
identify(IdentifyOptions(distinct_id=..., properties=...))	Set the current distinct ID + emit $identify.
set_distinct_id(id)	Set the ID without emitting $identify.
flush()	Force-send queued events. Always call before request returns on serverless.
shutdown()	Flush + stop the worker. Called automatically on process exit.

TrackOptions

@dataclass
class TrackOptions:
    distinct_id: str | None = None     # override the client's default
    timestamp: datetime | None = None  # override the auto-set UTC now
    context: EventContext | None = None

Decorators

The three decorators auto-applied by claude /init (the MCP slash-prompt). Each preserves your function signature; types and docs flow through @functools.wraps.

@growth_span(client, name, *, properties=None, distinct_id=None)

Emits <name>.started, <name>.completed (with duration_ms), or <name>.failed (with error_message, error_name). Works on sync and async functions automatically. Use on anything > 500ms or with non-trivial failure rate.

@growth_span(growth, "ai.generate", distinct_id=lambda req: req.user_id)
async def generate(req: GenerateRequest):
    return await anthropic.messages.create(...)

distinct_id may be a literal string or a callable that receives the wrapped function's args/kwargs and returns the ID.

@growth_step(client, funnel_step, *, properties=None, distinct_id=None)

Single-event funnel marker. Drop on whatever step represents progress through your activation.

@growth_step(growth, "onboarding.invited_team")
def invite_team(team_id: str):
    ...

growth_track(client, name, properties=None)

Imperative one-off track. Equivalent to client.track(name, properties), provided for symmetry with the decorator helpers.

growth_track(growth, "pricing_viewed", {"tier": "pro"})

Recommended event taxonomy

The MCP server's /diagnose, /weekly, and /pmf prompts know these names natively:

Stage	Events
Identity	signup_completed, login_completed, $identify
Activation	onboarding.started, onboarding.completed, first_<thing>_created
Revenue	checkout_started, checkout_completed, subscription_created, subscription_canceled, payment_failed
Performance	growth_span(growth, "checkout", ...), growth_span(growth, "ai.generate", ...)

Privacy

• Never put PII (email, raw IP, full address, payment details) in properties. Use distinct_id for identity. Backend ClickHouse never decrypts or displays properties as identity.
• ClickHouse TTL is 12 months by default — events older than that are dropped automatically.

Going deeper

The SDK is the ingest layer. The agent layer (Claude Code + MCP) is what makes growth-loop different — install the MCP server too:

claude mcp add growth-loop \
  -e GROWTH_API_KEY=pk_live_<your-key> \
  -e GROWTH_HOST=https://api.growth-loop.dev \
  -- npx -y @growth-loop/mcp-server

Then in Claude Code: /growth-init (auto-instrument), /growth-diagnose <metric>, /growth-weekly, /growth-pmf, /growth-icp, /growth-strategy, /growth-pivot.

(The growth- prefix is intentional — it keeps our prompts out of the way of Claude Code's built-in /init and any other editor's defaults.)

License

MIT