Recipes

The tool catalog lists every MCP tool. This page strings them into real tasks. Each recipe is a prompt you can paste into Claude, Cursor, or any connected client, plus the tools the agent will call and the point where it stops to ask you.

Build an onboarding flow

Go from an idea to a live native flow in one conversation.

Build a 4-screen onboarding flow for my app. Start with a welcome screen, then two value screens, then a notification-permission prompt. Use the setgreet tools. Show me the screens before you publish anything.

What the agent does:

1. list_apps (and switch_app) to set the app context so the flow is themed correctly.
2. compose_flow to generate the draft from your brief, or add_screen_from_pattern repeatedly for tighter control.
3. get_flow_context to show you the drafted screens.
4. Stops and asks before publish_flow, because publishing takes the flow live to your users immediately. The prompt above tells the agent to show you the screens first; make that a habit, since publish_flow is not one of the tools the server hard-gates behind an explicit confirm.

Iterate a flow by chat

Refine an existing draft without opening the editor.

In the flow "Welcome v2", change the paywall CTA copy to "Start my free week" and move the notification-permission screen to right after the welcome screen.

What the agent does:

1. list_flows to resolve the flow by name, then get_flow_context to read the current screens.
2. get_screen on the relevant screen, then update_component to change the CTA copy.
3. reorder_screens to move the permission screen.
4. Re-publish_flow when you confirm.

Edits target the DRAFT, so live users see nothing until you publish. If another teammate changed the screen since the agent read it, the tool returns conflict rather than overwriting -- ask the agent to re-read and retry.

Translate a flow into more languages

Localize an entire flow in one pass.

Translate the flow "Welcome v2" into Spanish, German, French, Portuguese, and Japanese.

What the agent does:

1. list_flows to resolve the flow, list_locales to see what already exists.
2. bulk_translate_flow once per target language -- each call translates every screen and preserves your {{personalization}} variables.
3. get_flow_localization to confirm coverage.

Run an A/B test and read the result

Test a change, then let the agent tell you which variant won.

Create an experiment on the flow "Welcome v2" with two variants: control, and a version where the paywall uses a free-trial layout. Split traffic 50/50 and start it.

What the agent does:

1. create_experiment with weighted variants (this creates a DRAFT).
2. start_experiment to move it to RUNNING.

Then, days later:

How is the "Welcome v2" paywall experiment doing? Is either variant winning yet?

3. get_experiment_results for per-variant metrics and statistical significance.
4. complete_experiment to declare a winner -- only after you confirm, since it is terminal and cannot be undone.

Audit a flow's drop-off

Find where users leave, before deciding what to fix.

Where are users dropping off in the flow "Welcome v2"? Break it down screen by screen and tell me which component loses the most people.

What the agent does:

1. get_flow_dashboard for the headline numbers (impressions, completions, drop-off).
2. get_flow_paths for the screen-by-screen funnel.
3. get_flow_components_analytics to find the specific button or input that loses users.
4. Optionally get_flow_inputs to read NPS, ratings, and form answers.

From here you can chain straight into "Iterate a flow by chat" to ship the fix.

Install the Setgreet SDK

Have your coding agent wire the SDK into your app.

The MCP server authors and publishes flows, but it runs against your Setgreet workspace, not your app's codebase. To install the SDK, point your coding agent at the Setgreet agent instructions -- a copy-paste block you add to your repo so Cursor, Claude Code, or Copilot installs and initializes the SDK correctly for your platform.

Next steps

• Tools -- the full catalog these recipes draw from.
• Setgreet for AI agents -- the drop-in instructions file for coding agents.
• Troubleshooting -- error codes and fixes.