The tools
One tool per question. The reading tools share three arguments:
range: a preset (today,yesterday,last_24h,last_7d,last_30d,last_90d,last_12mo,this_week,this_month) or{ "from": "2026-08-01", "to": "2026-08-31" }as dates in the site’s timezone, inclusive. Defaultlast_30d.filters: a list of{ "dimension", "op", "values" }, OR within a dimension and AND across, exactly as on the dashboard. Dimensions are the dashboard’s (path,entry_channel,country,device,browser,utm_campaign, … thesitetool lists them all) orprop:<key>for a custom event property.opisis,is_notorcontains.compare:previous_periodorprevious_year, where the answer has a “before”.
Every result is { "summary": "one sentence", "data": { … } }.
| Tool | Answers | Arguments beyond the shared ones | Returns |
|---|---|---|---|
site | what is this site | none | url, hostnames, timezone, goals with the KPI marked, and the ranges, dimensions and metrics the other tools accept. Call it first. |
summary | how is the site doing | compare | visitors, sessions, pageviews, bounce rate, engaged time, revenue; the previous period beside them |
timeseries | how did a metric move | metric, granularity (hour, day, week, month), compare | one value per bucket, the previous period’s beside it |
breakdown | which pages, sources, countries… | dimension, metrics (up to six), goal (an id, adds completions and conversion), limit (default 20, max 200) | ranked rows |
goals | are people converting | none | every goal with completions, converting sessions, conversion, revenue, median time to convert |
funnel | where people drop | steps: two to eight of { "kind": "any" }, { "kind": "pageview", "match": "/pricing" }, { "kind": "event", "match": "signup" } | sessions and share per step. The steps are given in the call; the funnels a site saved on the Goals screen are not yet listed here |
paths | how people reach an event | event, limit | the last pages before it, most common first |
attention | which pages hold and help | limit | attention share, minutes, read-through and influence per page (how they are defined) |
vitals | is it fast | limit | p75 per Core Web Vital with its band, and pages with their vitals |
realtime | who is here now | none | visitors in the last 30 minutes, pageviews per minute, pages, sources, countries |
bots | what crawlers fetch | family, limit, range only | hits by family, crawlers, pages, and the robots.txt, llms.txt and sitemap fetches (the families) |
accessibility | what got in the way of real visits, and what it cost | range from the presets yesterday, last_7d, last_30d, this_week, this_month only; filters on the four accessibility dimensions (input, zoom, motion, colors) plus at most one path and one device, each is one value; no compare | the measured count; each group (navigated by keyboard, keyboard with no pointer seen, browser zoom 200%+ and 400%+, reduced motion, forced colors) with its share, sessions, conversion and bounce with intervals, and its comparison; the findings by page and element with the criterion each relates to and the sessions that could have met it; what it cost per barrier; the notes and the privacy line. Nothing under 20 sessions or 5 completions; the answer changes once a day. Screen readers are not and cannot be counted (the whole page) |
notes | what happened when | range only | the notes in the range |
add_note | mark what I just did | text (up to 140 characters), at (ISO instant, default now) | the pinned note. Needs the notes scope; signed with the key’s name |
A wrong dimension, goal id or step is refused with a sentence naming what was expected, so an agent can correct itself without guessing. The accessibility tool refuses, rather than narrowing silently, any range outside its presets, a comparison, a second path or device filter, an is_not or contains, and the four accessibility dimensions are refused by every other tool; the reason names what the tool takes instead.