AI assistant
Numbers come from an analysis query run on this project; the assistant does not search a knowledge base or use vector retrieval. It does not create monitors or send notifications. The third tab is read-only MCP setup, documented in MCP for agents.
ConsoleAll-platform analyticsOperationsAI AssistantThe tabs are Ask, Why did it change, and MCP. If no model is configured, the corner badge reads Rules only (no model calls).
Metrics & definitions
Ask turns one sentence into a repeatable analysis: metric, time, comparison, breakdown, and filters. A question is parsed by rules first. When confidence is high and the turn is not a follow-up, the query runs with no model call. The model is used when confidence is low, no query was parsed, or the thread already has history. It must emit the same query shape, which is then executed. The reading is written from the result by rules. If a model is on and the site is under its daily cap, the model only rewrites that draft and must not invent numbers that are not in the result. If the model is missing or the daily cap is spent, the rule reading remains.
Why did it change explains how much each dimension value contributed versus the previous period or the same period last year. It can split channel, referrer domain, country, device, entry page, platform, and related dimensions, one or two levels deep. Contribution is that value's change divided by the total change.
How to use it
Type "visitor trend for the last 7 days versus the previous period" and press Enter. The answer includes the query. Open in analyzer and Save as insight are buttons you click; the assistant does not do them. Why did it change? carries the current metric into attribution.
Access & limits
Questions use the current project. Model calls have a per-site daily cap, shown at the top of the page. A cap of 0 means no cap. With no model configured, every answer stays on the rule path.
FAQ
What if the answer misses the question? Name the metric, time, and breakdown, or use a suggested question.
How are follow-ups handled? Follow-ups such as "now by country" or "mobile only" send the recent queries, and the model edits the previous one.
MCP for agents
The transport is Streamable HTTP with JSON responses and no server-sent stream. The server exposes query tools only, with no resources or prompt templates; natural-language answers follow AI assistant.
How to configure
On ConsoleAll-platform analyticsOperationsAI Assistant, open the MCP tab to copy the endpoint, client config, and curl sample. The path is /api/v1/mcp. Create the key under ConsoleConfigurationSettingsAPI keys with the Read stats scope (stats:read). The key is shown with a wck_ prefix only. A signed-in console session can call the endpoint as well.
Methods & tools
POST sends JSON-RPC. GET returns server info and tool names. DELETE ends a session; this server is stateless and returns success. Every tool except list_sites requires site_id.
| Tool | What it returns | Also required |
|---|---|---|
list_sites | Projects this credential can access | Nothing |
site_overview | Sessions, users, pageviews, and period change | A date range; identity is optional |
list_metrics | Built-in metrics, saved metrics, event names from the last 30 days, and dimensions | The project |
list_events | Event names from the last 30 days, by frequency | The project |
run_insight | A trend, table, funnel, or other analysis from a query | The project and the query |
list_saved_insights | Saved analyses | The project; optional name search |
run_saved_insight | One saved analysis | The project and the analysis id; the date range can be overridden |
ask | A plain-language question | The project and the question |
attribute_change | How dimension values contributed to a metric change | The project and the metric |
dashboard_summary | A summary of each dashboard card | The project; omit the dashboard to use the default |
Defaults & limits
The handshake accepts the protocol versions the server returns, including 2025-06-18. run_insight returns a compact result by default. Dates are a preset or from / to in the site timezone. Each tools/call checks access to site_id and counts toward open API usage and rate limits.
Example
The sample on the page is a POST to the endpoint with Authorization: Bearer and your key, method tools/call, tool ask, and arguments site_id plus a question. Replace the wck_ placeholder with your key.
Response & side effects
On success, the payload is text inside content, usually JSON. Tools are read-only: they do not write events, edit definitions, or create monitors. If ask cannot parse the question, it returns a clarification and does not run a query. If the project has no dashboard, dashboard_summary says so.
Errors & compatibility
Unknown tools, a missing site_id, and a key without the Read stats scope all fail. Notification methods return an empty body. Resource and prompt lists are empty. Do not assume other tools exist.