Example prompts
Copy any of these into a chat with your connected AI. Under each prompt: what the agent will do and which tools it uses. All work with the default connection unless marked otherwise.
Getting oriented
Section titled “Getting oriented”Show me what's in my SparkleTree workspace: which workspace am I connected to, what campaigns are live, and what's my top performer this week?The agent confirms scope with get_platform_info (organization), lists campaigns with query_campaigns, and pulls a performance summary with query_analytics — a complete first-run tour in one ask.
Import the product at https://example.com/products/aurora-lamp into SparkleTree and draft a campaign brief for it. Show me the brief before generating anything.The agent calls manage_products (import_from_url) to pull the product from the page, then drafts a brief and pauses for your approval before spending credits on generate_campaign.
How did my campaigns perform in the last 7 days? Give me Impressions, Reach, and clicks per campaign, tell me which variant is winning, and suggest one change.The agent runs query_analytics (snapshot, then cvg_dashboard for variant winners) and returns a short digest with a concrete recommendation — Impressions counted raw, Reach as distinct sessions.
What can you actually do in my workspace? List your tools and any recommended flows.The agent reads the mosaic://capabilities resource and get_platform_info (tool_profiles) and reports what’s enabled — useful right after connecting.
Campaigns
Section titled “Campaigns”Create a draft campaign called "Fall Warm-Up" for my Aurora Lamp product with two variants: one cozy/homey, one modern/minimal. Don't publish it.The agent uses manage_campaigns (create, then create_variant twice) and leaves the campaign unpublished for your review.
Generate a campaign for boosting weekend foot traffic. Walk me through the result before publishing.The agent confirms the credit spend, starts generate_campaign, polls query_campaigns (generation_status) until done, then presents the variants and waits before any publish.
Publish the "Fall Warm-Up" campaign and deploy it to my website surface.The agent looks up the campaign (query_campaigns) and surface (query_surfaces list_surfaces), confirms with you, then runs manage_campaigns publish and deploy.
Deploy my best campaign to Meta as an ad. Show me the plan first.Requires the meta_ops profile: reconnect with https://app.sparkletree.io/api/mcp?profile=meta_ops. The agent picks the winner via query_analytics, then uses the Meta tools to prepare the deploy and asks before anything goes live.
Products
Section titled “Products”List my products and flag any that are still drafts or missing an image — those can't be activated or placed by AI.The agent pages through query_products (list) and reports drafts and image gaps.
Update the description of my "Aurora Lamp" product to emphasize energy efficiency.The agent fetches the product (query_products get), proposes new copy, and applies it with manage_products (update) after you approve.
Analytics
Section titled “Analytics”Compare the variants in my "Fall Warm-Up" campaign. Is there a clear winner yet, or should I keep waiting?The agent uses query_analytics (cvg_dashboard, plus ab_test where relevant) and gives a keep-waiting-or-commit call.
Give me insights on my whole workspace: what's working, what's stalling, and where I'm leaving performance on the table.The agent runs query_analytics (insights, widening sections beyond the default summary only as needed) and returns a prioritized list.
What video templates do I have available?The agent lists them with query_video (templates) — a free read, no credits.
Render a short video from my best-performing campaign. Confirm the cost with me first, then keep me posted on progress.The agent finds the winner (query_analytics), confirms the credit spend, starts render_video, and polls query_video (progress) until the video is ready.