MCP examples

Real prompts for the unified analify server, once you've connected it — see
MCP for setup. Each one maps to specific tools your assistant can call
directly; you don't need to name them.

Profit and returns

"What's my real profit this month, after returns?"

Pulls Shopify revenue, ad spend across every connected platform, and returns
cost from your courier, and nets them against each other — not a single
platform's number.

"Which products are actually profitable after COGS?"

Calls shopify_get_profit_margin / shopify_get_profit_per_order against your
entered product costs. If coverage is low, ask first:

"How much of my catalogue has a cost on file?"

That's analify_get_cogs_coverage — worth checking before trusting a margin
figure. Low coverage means some products are quietly being priced as free.

Shipping and RTO

"Compare my carriers — who delivers better and who costs less?"

Runs bosta_get_success_rate and bosta_get_geographical_analysis
side by side. If you've connected Aramex or Mylerz too, ask it to compare all
three.

"How much cash is sitting with the courier right now?"

bosta_get_cash_cycles — the settlement-lag number, separate from what
Shopify shows as "sold".

"Which shipments are stuck and need a human today?"

Pulls anything not moving through the delivery lifecycle — the exception
queue, in plain language.

Ads, without the ROAS trap

"Which Meta campaigns look profitable but are actually losing money after
returns?"

Cross-references Meta's own reported ROAS against Bosta's return data — the
comparison no ad platform can make on its own, because it doesn't know what
came back.

"Why is TikTok wasting spend this week?"

Calls TikTok's reporting and creative-fatigue tools to find campaigns burning
budget with a falling result.

"Pause the campaign that's under our margin"

A write action — it changes a live campaign the moment the tool runs. Most
clients (Claude Code included) ask you to approve a tool call before it
executes; if yours doesn't, name the exact campaign and check its answer
before you ask it to act.

Investigating one order or customer

"What happened to order #4021 — where is it right now?"

Calls Shopify for the order, then Bosta for its tracking state, in one
answer instead of two tabs.

"Show me this customer's order history and their return rate"

Useful before deciding whether a repeat-return customer is worth serving COD
at all.

Catalog and creative

"List my Meta ad creatives that are fatiguing"

meta_ads_get_video_insights / creative tools flag declining performance
before spend gets wasted on a dead creative.

"Is my Shopify catalog in sync with what's on Meta?"

Compares shopify_list_products against the Meta catalog tools — catches
drift before it shows up as a broken ad.

Multi-step investigations

The strongest prompts chain several tools without you naming any of them:

"Give me a morning brief: yesterday's profit, any RTO spikes, and which
campaign needs attention today."

"I want to cut 20% of ad spend without hurting revenue — where should it
come from?"

"A customer is complaining their order never arrived — find it, check its
shipping status, and tell me what to say to them."

Getting better answers

• Say a real date range — "this week", "since the 1st", "last 30 days" —
  the assistant reads it the same way the dashboard does.
• Name the platform if you mean one — otherwise it reasons across every
  connected platform, which is usually what you want.
• Ask it to show the math — "why do you say that?" — every answer traces
  back to a real number, not a guess.
• Read your client's confirmation before approving a write. Analify's
  server executes the moment a write tool is called — your MCP client's own
  approval step, where it has one, is what stands between "asked" and "done".