> ## Documentation Index > Fetch the complete documentation index at: https://docs.abconvert.io/llms.txt > Use this file to discover all available pages before exploring further. # What Claude can do for you > Every operation Claude can perform on your behalf, grouped by intent. You don't need to memorize tool names — Claude picks them based on what you ask. Claude has access to 30 ABConvert tools. You don't call them by name — you describe what you want in natural language, and Claude picks the right one (or chains several together). This page lists what's possible so you know what to ask for. ## Discover & resolve These tools translate names you'd say in English into the IDs that ABConvert needs internally. You almost never trigger these directly — Claude calls them automatically when you mention a product, page, theme, etc. | What you say | What Claude does | | -------------------------------- | --------------------------------------------------------------- | | "the snowboards collection" | `list_shop_collections` → resolves to `/collections/snowboards` | | "the cotton hoodie" | `list_shop_products` → finds the right product GID | | "my sale page" | `list_shop_pages` → resolves to `/pages/sale` | | "the Refresh theme" | `list_shop_themes` → finds the theme by name | | "the alternate product template" | `list_shop_templates` → lists available Liquid templates | | "my US shipping zone" | `list_shipping_profiles` → returns zone IDs and current rates | You'll see Claude make these calls in passing. They never modify anything; they're read-only lookups against your Shopify catalog. ## Create a test Seven test types supported. Each one asks for a hypothesis, primary metric, and traffic split before creating anything. > "Test 10% off the Oxygen snowboard for US mobile visitors" > "Compare $39.99 vs $44.99 on the cotton hoodie, 50/50 split" > "Run a price test on all snowboards — minus 15% — but cap traffic at 50% of visitors" Claude resolves products by name, picks sensible defaults, asks you to confirm, and creates a draft. Variants can be percentage changes, fixed-amount changes, or absolute prices. [Price test details →](/experiments/price-test) > "Test free shipping over \$50 for US visitors" > "Bump my Express shipping rate from $9.99 to $14.99 in the US zone, 50/50 split" > "Test removing the Standard shipping option in the Canada zone — see what happens to AOV" Modifies, hides, or adds rates per zone. Untouched rates are auto-preserved (you don't need to enumerate every existing rate). [Shipping test details →](/experiments/shipping-test) > "A/B test my Dawn theme vs the Refresh theme, 50/50" > "Test the new theme I just uploaded against the live one" Claude resolves theme names. Path-condition routing (testing only on certain URLs) isn't yet exposed via MCP — use the wizard for that. [Theme test details →](/experiments/theme-test) > "Test the alternate product template I created" > "Compare two product page layouts: the default and the 'minimal' template" Picks alternate Liquid templates by name. Available for any template type (product, index, collection, page, blog, article, cart, search). [Template test details →](/experiments/template-test) > "Send 50% of /products/old-hoodie traffic to /products/new-hoodie" > "Test a redirect from /sale to /pages/black-friday for US mobile visitors" Pure URL composition — Claude resolves friendly names ("the sale page", "snowboards collection") to real paths. Supports exact match, regex, contains, starts-with. [URL redirect test details →](/experiments/url-redirect-test) > "Test 15% off the whole order against no discount, 50/50" > "Compare free shipping over $50 vs a flat $10 off on all snowboards" > "Test a volume discount — buy 2 get 10% off — on the cotton hoodie" Builds discount variants against a no-offer control: percentage or fixed amount off products, the order, or shipping, plus volume and threshold tiers. Claude resolves products and collections by name. [Offer test details →](/experiments/offer-test) > "Rename my 'Test rate 1' delivery option to 'Standard Shipping' at checkout" > "Hide the slowest delivery method for US orders over \$100" > "Reorder payment methods so Shop Pay shows first" Plus-only. Hides, renames, or reorders delivery and payment methods, or adds checkout UI blocks, against an unchanged-checkout control. [Checkout test details →](/experiments/checkout-test) ## Lifecycle: launch, preview, pause, resume, end Once a test exists (whether you created it via MCP or in the wizard), you can drive it from chat: | What you say | What happens | | ------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | "Launch the Oxygen test" | Goes to Active and real traffic starts flowing — from a Draft, or straight from a Preview once you're happy with it. | | "Preview it first" | Goes to Preview status; Claude returns a URL you can open in your browser to QA the variant before launching. Share with the team if you want a second pair of eyes. When you're ready, "launch it" goes live; "put it back to draft" reopens it for edits. | | "Put it back to draft" | Pulls a previewed test back to Draft so you can change its setup before launching. | | "Pause the snowboard test" | Stops assigning new visitors. Existing data preserved. | | "Resume it" | Goes back to Active. | | "End the test, the variant won" | Closes it permanently. Data stays for analysis; new visitors stop being assigned. | | "Clone the Oxygen test" | Duplicates it as a new draft you can tweak before launching. | | "Update the hypothesis to ..." | Patches metadata on a draft (name, description, hypothesis, primary metric). | Deleting a test isn't available from chat — the API token can't delete. To remove a draft, end it (if it ever launched) or delete it from the dashboard. Active tests must be ended first either way. ## Analytics: query results You don't need to open the dashboard to check on tests: | What you say | What Claude returns | | ------------------------------------------------------- | ----------------------------------------------------------------------------------------- | | "How is my Oxygen price test doing?" | Per-variant metrics: views, conversions, AOV, RPV, p-value, confidence interval | | "Show me the breakdown by device" | Same metrics, broken out by mobile / desktop / tablet | | "Compare the Oxygen and Hydrogen tests" | Side-by-side comparison of both tests' winners and lifts | | "Which variant is winning?" | Plain-language summary with statistical-significance call-out | | "Is the snowboards test statistically significant yet?" | Yes/no answer based on the current sample size | | "Which tests are ready to act on?" | Active tests that have cleared significance and run long enough to call | | "When did we launch this? Why is it paused?" | The test's lifecycle history — created, launched, paused, resumed, ended, with timestamps | Available breakdown dimensions: date, device, country, traffic channel, traffic platform, visitor type, landing page, referral domain, and each UTM parameter (source, medium, campaign, term, content). Page-experience tests add page URL, template, and click destination. Pass up to two at once (e.g. device × country). ## What's NOT supported (yet) Still wizard-only: * **Visual editor tests** (DOM mutations like text/image/link changes) — Claude can't see your storefront the way the visual editor's click-to-record can. * **Path-condition routing on theme tests** (testing only on certain URLs) — create the test in chat, then add path conditions in the wizard. We'll add these as merchant demand drives priority. If you want a specific one prioritized, let support know. ## What's NOT supported, period Some operations are intentionally not exposed: * **Billing/plan changes** — these go through Shopify, not the API. * **Shop-level settings** that aren't experiment-scoped (CSS, theme injection points, etc.) — wizard only. * **Deleting tests** — the API token can't delete. End a running test from chat; remove drafts from the dashboard.