Claude Skill · Open Source · MIT

SiteSplit

A/B testing for Shopify that you run by talking to Claude. Zero servers, zero signup — everything lives in your Shopify, controlled by your Claude.

"Create a test." "What's winning?" "Ship the winner." That's the whole interface.

The dashboard — generated locally, no server
Live sample output. Claude builds this HTML on your machine from your store's real funnel data — sessions → add-to-cart → checkout → purchase, per variant.
How it works
01

Say the test

"Test /pages/a against /pages/b, 50/50." Claude saves it to a Shopify metaobject — your store is the database.

02

Paste one block

A tiny snippet goes in your theme's <head>. Sticky 30-day bucketing, redirect before paint — no flicker.

03

Orders self-report

The snippet tags the cart, Shopify carries the tag onto every order. Revenue attribution with no pixel server.

04

Honest reads

Claude pulls your real funnel from Shopify analytics and tells you "leaning variant" or "too early" — never a fake confidence %.

Why it's different

Conversation-first

Every other A/B tool is a dashboard you click through. SiteSplit is a skill your Claude runs — create, launch, analyze, and ship winners in plain English.

Your data never leaves your store

No third-party script domain, no analytics SaaS in the middle. Config lives in metaobjects, results live in your orders and your Shopify analytics.

Full funnel, free

Sessions → add-to-cart → reached checkout → purchased, per variant, straight from Shopify's own analytics. Plus revenue per variant from order attributes.

Optional: Microsoft Clarity layer

Connect the free Clarity MCP and the dashboard adds scroll depth, rage clicks, and dead clicks per variant — the why behind the winner.

Get it

Drop it into your Claude skills folder, connect the Shopify connector, and say "create an a/b test."

↓ Download the skill View source on GitHub →
# install (Claude Code)
unzip sitesplit-skill.zip -d ~/.claude/skills/
# connect Shopify: Claude Settings → Connectors → Shopify
# then just say:
"create an a/b test between /pages/a and /pages/b, 50/50"