What is MCP?
The Model Context Protocol (MCP) is an open standard created by Anthropic. It lets AI assistants connect securely to external data sources and tools. The Shogun MCP Server uses this protocol to give AI assistants controlled access to:
Your Shogun test details and analytics
Store-wide analytics collected by Shogun, filterable by date, device, traffic source and ad campaign
Test actions such as creating, starting, pausing and stopping tests
Your Shopify store's products, collections, pages, templates, themes and audiences
Requirements
An active Shogun A/B Testing account
A Shogun A/B Testing plan that includes MCP access (Advanced plan, Unlimited plan, or Custom Contracts)
An MCP-compatible AI client, such as Claude Desktop, Claude Code, ChatGPT or Cursor
Installation
Shogun hosts the MCP server, so there's nothing to install. The server URL is:
https://ai.shgapi.com/mcp
Connecting Claude
Free Claude plans allow one custom connector. Additional custom connectors require a Claude Pro, Max, Team, or Enterprise plan.
Claude Desktop
1. Open Connectors
Open Claude Desktop and go to Settings > Connectors.
2. Add a custom connector
Click Add, then Add custom connector.
3. Enter the server details
Set the name to "Shogun" and the URL to https://ai.shgapi.com/mcp. Leave the OAuth settings empty, then click Add.
4. Connect
Click Connect to continue with authentication.
5. Sign in
Enter the email address associated with your Shogun account. Click Sign in in the confirmation email, then Authorize on the final screen. You'll be returned to Claude Desktop, connected.
6. Enable the connector in your chat
When you start a chat, make sure the Shogun connector is enabled.
Claude Code
Add the Shogun MCP Server using the CLI:
claude mcp add --transport http shogun https://ai.shgapi.com/mcp
The first time you use the server, Claude Code opens a browser window to complete sign-in. Tokens are stored securely and refresh automatically.
Alternatively, add it to your .mcp.json file. Put it in your project root for one project, or in ~/.claude.json for user-wide access:
{
"mcpServers": {
"shogun": {
"type": "http",
"url": "https://ai.shgapi.com/mcp"
}
}
}
Connecting ChatGPT
ChatGPT doesn't always interpret the data we return as reliably as Claude does. For the most accurate results, we recommend connecting Claude instead.
Custom connectors require a ChatGPT Plus, Pro, Team, or Enterprise plan with Developer mode enabled.
In ChatGPT, click your profile name and go to Settings > Apps > Advanced settings.
Turn on Developer mode.
Click Create app.
Set the name to "Shogun" and the URL to
https://ai.shgapi.com/mcp. Leave the OAuth settings empty. Check I understand and want to continue, then click Create.Enter the email address associated with your Shogun account. Click Sign in in the confirmation email, then Authorize on the final screen. You'll be returned to ChatGPT, connected.
Using it in a chat
Start a new chat and click the + icon.
Select More > Developer mode.
Choose Add sources and enable Shogun.
Connecting Gemini
Custom connectors require a Google Gemini Enterprise plan.
Add the Shogun MCP Server using the Gemini CLI:
gemini mcp add --transport http shogun https://ai.shgapi.com/mcp
The first time you use Shogun, Gemini opens a browser window to complete sign-in.
Alternatively, add it to ~/.gemini/settings.json for user-wide access, or to .gemini/settings.json for a single project:
{
"mcpServers": {
"shogun": {
"httpUrl": "https://ai.shgapi.com/mcp"
}
}
}To manage the connection from the CLI:
gemini mcp list # View all configured servers
gemini mcp remove shogun # Remove the server
Connecting other clients
Any MCP-compatible client, including Cursor, Windsurf and Copilot, can connect to Shogun over HTTP. Add a custom or remote MCP server using this URL:
https://ai.shgapi.com/mcp
Most clients support automatic OAuth discovery. Provide the URL above, then complete the sign-in flow when prompted, using the email address associated with your Shogun account. Your client's documentation has its specific steps for adding a remote MCP server.
Multi-store support
When you authenticate, you get access to every Shopify store associated with your Shogun email address.
Each store needs a plan that includes MCP access. You can switch between stores through the MCP and work across several stores in a single chat.
To always work with a specific store, add it to a system prompt or custom instructions in your AI client.
Filtering your analytics
The MCP uses the same filters as the Insights tab in the Shogun app. When you ask about a test or your store in chat, you'll get the same numbers you'd see with those filters applied in the app. Just describe what you want and your assistant applies the right filters.
Date range: any range up to 365 days. If you don't give one, store analytics cover the last 30 days, and test results cover the test's full run. Dates follow your store's time zone, and both the start and end days are included.
Device: desktop or mobile.
Traffic source: Google Ads, Meta Ads, paid social, organic social, organic search, AI agents, referral, email, or direct.
Ad campaigns: specific campaigns or ad groups. Your assistant can't look these up yet, so include the campaign or ad group ID from Google Ads or Meta Ads Manager in your question. See Coming soon below.
Period: weekly or monthly breakdowns of store-wide KPIs.
Filtering a test's results
Filters narrow the metrics you see for each variant, such as conversion rate, AOV, and revenue per visitor. The test's confidence level and winner are always calculated on the whole test, so filtering to one traffic source won't change whether a test has reached significance.
Smaller numbers
Filtering to a single campaign or device can leave a much smaller group of visitors. With fewer visitors, percentages can swing sharply on very few orders, so check the session and order counts before reading too much into a large difference.
Coming soon: browsing your campaigns and ads
Today, your assistant can filter results to Google Ads or Meta Ads traffic. It can also narrow to a campaign or ad group, but only if you give it the ID. We're adding a tool that lets your assistant list the campaigns, ad groups, and ads Shogun has synced for your store. Once it's available, you'll be able to:
Refer to campaigns by name, for example, "How did my PDP test do for the Autumn Sale campaign?" without looking up IDs.
Drill down from a campaign into its ad groups and individual ads to see which ones send visitors who convert.
Ask your assistant to compare ads against each other within a test.
We'll update this article when it's live.
Examples & Best Practices
Use Cases
The Shogun MCP has full access to your A/B Testing environment. It's designed to speed up your test roadmap through better test analysis and automated testing workflows. The full list of capabilities is in the Available Tools section below. Here are some examples of what you can ask:
"How are my running A/B tests performing?"
"How did my PDP template test perform for mobile visitors from Meta Ads over the last two weeks?"
"Compare conversion rate and AOV for email vs. organic search traffic this month."
"Show me monthly revenue and conversion rate for the last six months."
"List the Shopify order IDs attributed to each variant in my PDP test, so I can check them against my Shopify order export."
"For any running tests that have reached 90% confidence in a winner, promote that winner and end the test."
"Analyze my store analytics and recommend a few pages to test next."
"Create a template test. Set the variant to my 'pdp-review-style-variant' template, set the goal to Add to Cart Rate and the traffic split to 50/50, and target mobile visitors only."
"Create a page test on my cart page using my 'cart-free-shipping-bar' template as the variant, with Conversion Rate as the goal."
"Create a summary of all tests running across Store A, Store B, and Store C."
Creating tests through the MCP
Tests start as drafts. A new test gets no traffic until you ask your assistant to start it, and you confirm. You can open a draft in the Shogun app straight away to review it, including previewing each variant with its preview link.
Audiences. Your assistant can target a test at a built-in audience, such as mobile visitors, returning visitors or logged-in customers, or at any audience you've saved in the Shogun app.
Page tests work on any page. A page test can target a single product, collection, page, blog, blog post or metaobject page. It can also target your homepage, your cart page, and the page that lists all your collections (
/collections). Just tell your assistant which page you mean.Price and Product Details Tests need theme prep first. Complete theme preparation in the Shogun app and approve the prepared theme before you ask your assistant to create one of these tests. The MCP can reuse an approved theme but can't start theme prep itself. Starting one of these tests publishes the prepared theme as your live theme, and stopping the test doesn't undo that. Product Details Tests also need a recent theme prep: if your assistant says your approved prep is too old, rerun theme preparation in the Shogun app first.
Test type can't be changed later. To switch test types, ask your assistant to create a new test instead.
Best Practices
Approving each read-only tool call by hand can get cumbersome, so we recommend setting read-only tools to Always allow.
We recommend keeping write/delete tools set to Needs approval, so no action is taken on your storefront without your approval. Only switch these to Always allow once you're comfortable working with Shogun's MCP.
Security
Agents can only access the Shopify stores your account has access to.
By default, any action an agent takes requires your approval before it goes ahead.
Every write action an agent takes is recorded in that test's Audit Log.
Shopper personal information is never returned. Order data is aggregated and stripped of PII.
Available Tools
Store Management
list_stores
Lists every store your Shogun account can access, and shows which one is active. For each store, it returns the name, .myshopify.com domain, primary domain, currency, time zone, and whether A/B Testing is active.
switch_store
Changes which store your later questions apply to. You only need this if you have multiple stores. The store must be linked to your account and have A/B Testing active.
Experiment Tools
list_experiments
Lists the A/B tests on the active store. Your assistant can filter the list by status, test type, name and date range, and page through long lists.
get_experiment_details
Returns a test's full setup: name, status, test type, goal metric, variants with traffic split, notes, timeline, and configuration. Sensitive keys are never included.
get_experiment_results
Returns each tracked metric (label, goal, current value, and improvement) overall and per variant.
Supports analytics filters.
get_experiment_statistics
Returns the Bayesian statistics behind a ship/kill decision. This includes whether the result is statistically significant, the confidence level, improvement, the winning and losing variants, win probabilities, and session counts.
Supports analytics filters. Filters narrow the reported metrics only; significance always reflects the whole test.
get_experiment_orders
Returns the orders attributed to a test, grouped by variant and device, with the order count, revenue and profit for each group. Ask for order IDs and it also lists the Shopify order ID of every attributed order, which is useful for matching Shogun's numbers against a Shopify order export. Filterable by date. Shopper personal information is never included.
Analytics Tools
get_business_kpis
Returns store-wide KPIs, KPIs for traffic in active tests, and a recent A/B testing performance summary. It's useful for "How is my store doing?" questions and weekly reports.
Supports analytics filters. Includes weekly or monthly periods.
get_product_analytics
Returns per-product revenue, add-to-carts, cart conversion rate, and related metrics.
Supports analytics filters.
get_test_recommendations
Ranks products by testing opportunity, based on revenue and cart-conversion gaps. It also shows whether each product is already being tested, and suggests a test type for each.
Supports analytics filters.
Catalog Tools
get_products
Lists Shopify products with their ID, title, handle, template, and variants with prices.
list_collections
Lists Shopify collections with their ID, title, handle, and template.
list_pages
Lists Shopify pages with their ID, title, handle, and template.
list_themes
Lists the Shopify themes installed on the active store.
list_templates
Lists the template files in the active store's live theme, each split into a template name and suffix.
list_audiences
Lists the audiences a test can target. These include built-in audiences, such as mobile, desktop, new or returning visitors, and logged-in or logged-out customers, plus any custom audiences you've saved in the Shogun app.
get_shogun_schema
Lets your assistant look up Shogun's conventions, so it can answer accurately and set up tests correctly. That includes the supported test types (page, template, theme, split URL, price, product details, and checkout), what each test type needs, the available analytics filters, and safety guarantees. It takes no parameters.
Management Tools
create_experiment
Creates a draft test. Because a draft gets no traffic, your assistant doesn't need to ask you to confirm it in chat, though your AI client may still ask you to approve the action.
update_experiment
Updates a draft test. Like creating a draft, this doesn't need your confirmation in chat.
start_experiment
Starts a draft test. Requires your confirmation.
pause_experiment
Pauses a running test. Requires your confirmation.
resume_experiment
Resumes a paused test. Requires your confirmation.
select_winner_and_stop
Selects a winner and stops a test. Requires your confirmation. This is the only way to stop a test, because choosing a winner is always an explicit decision. To stop without keeping a variant, select the original as the winner. For page and template tests, your assistant can also apply the winning template and remove the losing variants' templates.
test_again
Creates a new draft from a finished test. Requires your confirmation.
add_experiment_note
Creates or updates a test's notes. Like creating and updating drafts, this doesn't need your confirmation, since it never affects a live test.