Get reliable answers

Once MCP is connected and you've asked a few questions, this page helps you trust what comes back. It covers the settings that change your numbers, how to get answers faster, how to check them, and when another tool is the better fit.

Defaults and limits

Most "that number looks wrong" moments come from a default you didn't know about. Here's what MCP does unless you say otherwise:

Get faster answers

MCP is built for back-and-forth questions. A few habits keep answers quick.

Narrow your date range. The shorter the window, the faster the answer:

• 30 days: your default for exploring
• 90 days: for quarterly trends
• 365 days: for year over year, but expect longer waits

Start small and widen the range once you know which question matters.

Be specific. Asking about 20 specific ASINs is faster than asking about "all my ASINs". Same for "brand X" vs. "all brands". The more filters you give, the less DataHawk has to scan.

Stick to one topic per conversation. Broad prompts like "do a complete audit of my business" can trigger 20+ tool calls and time out. Use one conversation for sales, one for advertising, one for inventory. You'll get cleaner answers and won't hit rate limits.

Verify your numbers

Ask to see the tool call. For any number that matters, ask the AI to show the DataHawk tool call it ran. You'll see the exact filters, dates, and data used. See How to iterate when the answer looks wrong.

Cross-check against your dashboards or data warehouse. When a number really matters (a board meeting, a supplier call), run the same date range and marketplace in your dashboards, Snowflake, or BigQuery and compare. If they differ, check the defaults first.

Reconcile with a screenshot. Claude and ChatGPT both accept images. Drop in a Seller Central screenshot and ask:

"Reconcile this against DataHawk data for the same period."

Work across marketplaces, platforms, and workspaces

Always name the platform and marketplace. MCP includes every connected marketplace by default, which is rarely what you want. Say it explicitly:

• "For Amazon US only..."
• "Across all my European Amazon marketplaces..."
• "On Walmart, last 30 days..." (advertising and keyword ranks only for now, see Walmart coverage)
• "Both Amazon and Walmart combined..."

Five extra seconds in the prompt saves five minutes of reconciling later.

Compare marketplaces in one prompt. Ask for a side-by-side view:

"Compare revenue, units, ACOS, and conversion across my US and UK marketplaces, last 30 days."

When the marketplaces use different currencies, totals come back in USD. Add "in local currency" to see each marketplace in its own currency.

Connect each workspace separately. If you run separate DataHawk workspaces for different brands or clients, connect each one to your AI client. One conversation can't span several workspaces yet.

Investigate "why" questions

Questions that start with "why", "what's driving", or "what changed" trigger a dedicated investigation flow. For example: "Why did revenue drop?" or "What's driving the ACOS spike?" The AI builds hypotheses, gathers evidence across domains, and explains its reasoning instead of just listing numbers.

Redirect it if it goes the wrong way. Course-correct in plain English:

"Skip the catalog analysis. Focus on advertising changes."

If a conclusion still feels wrong:

1. Ask for the raw data: "Show me the raw sales data for those 7 days, day by day."
2. Check the same numbers in your dashboards or data warehouse.
3. Email [email protected] if the gap remains.

Pick the right tool

MCP is great for exploring and for first drafts: generate a weekly summary, then polish it in a Google Doc or wherever you usually write. For some jobs, another tool fits better:

Not sure which to pick? Email [email protected] with what you're trying to do, and we'll point you to the fastest path.

---

What's next