Skip to main content

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.

The MCP integration is in beta. Reach out to support to opt in.
ABConvert exposes a Model Context Protocol (MCP) server that turns Claude (or any MCP-aware AI assistant) into a working interface for your A/B testing program. Instead of opening the dashboard, picking a test type, filling out a wizard, and clicking through five screens to launch — you describe what you want in plain language and Claude does the rest.

​
What you can do

You:    Test 10% off the snowboards collection for US mobile visitors
Claude: I'll set this up. Hypothesis: "10% discount lifts revenue per visitor among
        US mobile shoppers", goal: revenue per visitor, 50/50 split, 100% traffic,
        audience: US visitors on mobile. I'll save it as a draft so you can review.
        Tweak any of these or say "create it" to proceed.
You:    create it
Claude: Draft created. Want me to launch now or generate a preview link first?
That’s a real session. Claude resolves “the snowboards collection” to the right Shopify GID, picks sensible defaults, asks for confirmation, and creates the test on your store.

​
Why use it

Faster than the wizard

For tests you’ve run before, going from idea to draft is one paragraph instead of clicking through six screens.

Always-on assistant

Ask Claude “how is my hoodie price test doing” — get the metrics, the breakdown by device, the cross-test comparison. No dashboard navigation.

Works with the team

Anyone on your team with the ABConvert MCP token can drive tests from their preferred AI assistant. Same surface across Claude Desktop, Claude Code, OpenClaw, Fin.

Programmatic without code

All the power of an API without writing any code yourself — Claude does the API calls, you supervise.

​
When NOT to use it

MCP is great for:
  • Repeating tests you’ve run before with small variations (“same as last week but for the Hydrogen instead of Oxygen”)
  • Querying analytics without context-switching to the dashboard
  • Bulk lifecycle work (pause five tests, end three others, start a draft) in one chat
The wizard is still the right tool for:
  • First-time test design where you want visual guidance and live previews of what the variant will look like
  • Complex audience targeting setups that benefit from the visual filter builder
  • Visual editor tests (DOM mutations) — those need a real click-to-record interface; MCP doesn’t help here

​
What’s supported

Price test

Create, preview, launch, end. Resolve products by name. Apply percentage or fixed-amount changes.

Shipping test

Modify rates, hide rates, add new rates per zone. Auto-preserves untouched rates.

Theme test

Compare two themes by GID. Resolve theme names from Claude.

Template test

Test alternate Liquid templates per type (product, index, collection, page).

URL redirect test

Send traffic to different landing pages. Resolve URLs from natural language.

Analytics

Query results, breakdowns by device/country, cross-test comparison.
Offer, checkout, and visual editor tests aren’t yet exposed via MCP — use the wizard for those.

​
How it works

You mint an API token from ABConvert’s settings. You paste it into your AI assistant’s MCP config. Your AI assistant then talks to https://mcp.abconvert.io/mcp over HTTPS using that token. The MCP server validates the token, scopes everything to your shop, and forwards the work to ABConvert’s backend — the same backend the wizard uses.
MCP is currently available on the Growth plan and above. Pricing →

​
Next steps

Connect in 5 minutes

Step-by-step setup for Claude Desktop and Claude Code.

What you can ask

The full list of operations Claude can perform on your behalf.