Use Playbook from Claude or ChatGPT

Playbook ships an MCP connector — a secure way to plug your store into Claude, ChatGPT, or any AI assistant that supports connectors. Once it's linked, you can manage Playbook by chatting: ask how your tests are doing, get ideas grounded in your real brand and catalog, and create and launch experiences without opening the app.

What you can do

See what's working. Ask about a single experience or a summary across your recent experiences (up to the 50 most recent) and Playbook answers with the real numbers — impressions, conversions, conversion-rate and revenue lift, "revenue won," and Bayesian win probability. It can also list your experiences and tests, your plays, your brand voice and design, your best-selling products, and the built-in templates.

Get grounded ideas. Ask "what experiences should I create?" or "what plays should I run?" and Playbook suggests them — each grounded in your actual brand voice and best-sellers, with a rationale and a ready-to-use prompt. These aren't generic tips; they're specific to your store.

Create and launch. With write access, the assistant can create a play, create a fully AI-generated experience (it drafts the sections, copy, and imagery and hands back an editor link), move an experience into a play, and change an experience's status — take it live, pause it, or end it.

Under the hood these are named tools (list_experiences, get_performance_summary, suggest_experiences, create_experience, activate_experience, and more) — but you never call them by name. You just ask.

Connect it

1. In Claude or ChatGPT, open the add-a-connector flow (in Claude, custom connectors; in ChatGPT, a custom MCP connector).
2. Paste the Playbook MCP URL: https://www.heyplaybook.com/api/mcp
3. You'll be sent to Playbook to sign in. Then you'll see a consent screen headed "Connect [app] to Playbook." It shows:
   • Which app is connecting and where it redirects — check the redirect host matches the app you meant to connect.
   • The permissions you're granting (viewing your data, and — if requested — creating and launching).
   • A store picker, if your account has more than one store. The connection is bound to the single store you choose.
4. Approve, and the assistant is connected to that store.

You can disconnect at any time, and the assistant never sees your password — it only gets the access you approve on that screen.

What it can and can't touch

• One store, scoped. A connection is tied to the single store you pick at consent. It can't reach any other store.
• Two permission levels. Read covers analytics and ideas; write covers creating, moving, and starting/pausing/ending tests.
• Same plan rules as the app. Reading is always available. Taking an experience live requires an active Playbook plan, exactly like in the app — but pausing or ending a test is never blocked, so you can always stop serving.
• Newer experiences, not legacy tests. Analytics and the create / move / launch tools work with your new-model experiences. Older A/B tests still appear when you ask the assistant to list everything, but they can't be summarized, analyzed, moved, or changed through it.
• No destructive or account changes. There's no delete, and it can't touch billing, your team, or your brand settings.

Good things to ask

• "Which store am I connected to?" — a quick way to confirm the scope before you make any changes.
• "How are my experiences doing this week?"
• "What experiences should I create for first-time gift shoppers?"
• "Create an experience for my bestseller aimed at returning customers, and file it under a Holiday play."
• "Pause the losing test on my hydration collection."

Next: speed up the building itself — Save sections and templates for reuse.