Prompts
The MCP server ships five analytics workflows that chain the right tools in the right order, then ask the agent to present results in a specific shape. Invoke them from any client that supports MCP prompts.
Available prompts
| Prompt | What it does | Arguments |
|---|---|---|
weekly_report | Weekly summary: headline metrics with week-over-week deltas, top pages, referrers, countries, daily trend, and one thing to watch. | none |
traffic_diagnosis | Why did traffic change? Walks channel, country, device, and page breakdowns to isolate what moved and by how much. | period (optional, default 7d) |
conversion_audit | Finds the biggest drop-off in your funnels, then segments by device, country, and channel to tell you who converts worst. | funnel_name (optional) |
channel_breakdown | Compares channel quality by bounce rate, session duration, and custom event rate. Volume vs quality for each source. | period (optional, default 30d) |
page_performance | Single-page deep-dive: trend, referrers, devices, and engagement vs site baseline. | pathname (required), period (optional) |
Invoking prompts
In Claude Desktop and Claude Code, prompts appear as slash commands (e.g. /weekly_report). In Cursor and VS Code, trigger them from the MCP prompts picker. Any MCP client can call them by their prompt name and argument map.
Prompts assemble a single user-role message that tells the agent which tools to call and in what order. You can always skip the prompt and call tools directly. See Examples for the patterns they encode.