Skip to main content
Copy any of these into your MCP-connected AI client for quick, one-off questions — just replace the placeholders in brackets.
For repeatable workflows, use Agent Skills instead. Skills auto-activate and carry the judgment for a whole task — period reviews, tracking-link retrieval, conversion scrubbing, and more — without copy-pasting prompts. The starter prompts below are best for ad-hoc, exploratory queries; for end-to-end tool traces, see Agentic Examples.
  • Be specific about date ranges. The agent performs better with explicit dates (2026-04-01 to 2026-04-30) than relative terms like “last month.”
  • Ask for one thing at a time in complex investigations. Multi-step workflows are more reliable when each step is confirmed before the next.
  • Use IDs when you have them. Filtering by offer_id or affiliate_id is faster and more accurate than filtering by name.
  • Specify currency and timezone if your network has non-USD defaults or you’re comparing across regions.
  • Paginate explicitly for large datasets. Ask the agent to “get the next page” or set page_size in your prompt when working through long lists.
Start every new session with get_account_info or ask the agent to call it automatically. This confirms your default currency and timezone, which affects how reports and timestamps are interpreted. The tool needs the Accounts permission and is hidden from a key without it (for example a Read Only Reporting key); reports default to the network’s timezone and currency regardless.

​
Starter prompts

Confirm your session context
“Call get_account_info and summarize my network name, default timezone, default currency, and whether I have limited affiliate scope.”
Performance snapshot
“Give me a performance summary for yesterday — total clicks, conversions, revenue, payout, and profit — in my network’s default timezone.”
Top offers this month
“What are my top 10 offers by revenue this month? Show name, clicks, conversions, CVR, revenue, and profit, sorted by revenue descending.”
Compare this week to last week
“Compare this week’s performance to last week across clicks, conversions, revenue, and profit. What changed significantly?”
Biggest movers by offer
“Which offers had the biggest change in EPC this week vs last week? Rank by the EPC delta and show each offer’s current vs prior value.”
Diagnose a click or conversion
“Look up transaction ID [TRANSACTION_ID] — offer, affiliate, timestamp, unique flag, the error code and its meaning, and whether it converted.” (or: “Look up conversion [CONVERSION_ID] — offer, affiliate, advertiser, payout, status, error code, click timestamp, and sub/adv parameters.”)
Why are clicks invalid / conversions rejected
“Run a performance report for affiliate [AFFILIATE_ID] between [FROM] and [TO] with dimensions=click_error_code. Show each reason and its count, sorted descending. Ignore error code 0.”
To restrict an error-code breakdown to one reason, use the filter error_code:<code>.
Unpaid invoices for an affiliate
“List all invoices for affiliate [AFFILIATE_ID] with status unpaid. Show billed amount, the period, and whether it’s payable.”
The invoice entity type needs no filter: status="unpaid" alone lists every unpaid invoice network-wide. To scope to partners, pass affiliate_id (one id, or up to 100 joined with |); the ids actually applied are echoed as applied_affiliate_ids.

​
Registered prompt resources

Three registered prompts carry the same recipes as the skills (compare-periods matches like-for-like-comparison, network-movers keeps its name, and manager-book matches manager-book-review), for clients that discover prompts through the protocol instead of installing skills. Arguments marked * are required; dates are YYYY-MM-DD.

​
Registered prompt resources (legacy)

The first four are superseded by Agent Skills. The MCP Server historically registered a few named prompts — troubleshoot-click, partner-health-check, revenue-analysis, and offer-launch-readiness — invokable by name via the MCP prompts/get method. Skills now cover these same workflows with auto-activation and richer judgment, so the registered prompts are retained only for backward compatibility with clients that have a native prompt browser. They are still returned by prompts/list, but they are not recommended for new integrations and may be removed in a future release — use Skills or the starter prompts above.
For clients that do call them, these are the registered names and arguments (* = required):