Documentation

Getting started

OverviewQuickstartConnect ClaudeConnect ChatGPTConnect Cursor, Claude Code, VS Code

Core concepts

How it worksWhat gets collectedData freshnessMetrics and moneyGetting good answers

Reference

All tools37
Start here
get_cro_overviewget_opportunity_sizingget_anomalies
Funnel and conversion
get_funnel_analysisget_conversion_correlationsget_cart_analysisget_exit_analysisget_journey_patternsget_session_insights
Page and element diagnostics
get_top_issuesget_element_click_analysisget_page_dom_snapshotget_click_heatmapget_click_analysisget_page_vitalsget_page_performanceget_form_analyticsget_hover_analysisget_section_visibilityget_js_errorsget_page_templates
Products and discovery
get_product_performanceget_product_catalogget_product_changesget_search_analysisget_product_page_structure
Individual sessions
get_session_journeysget_session_summaries
Measuring a change
get_before_after
Custom events
get_custom_event_suggestionstest_custom_event_rulecreate_custom_event_rulelist_custom_event_rulesdelete_custom_event_ruleget_custom_event_definitionsget_custom_event_statsget_custom_events

Guides

Playbooks21Custom eventsPlans and limits

Support

TroubleshootingPrivacy and security

get_session_summaries

AI-written narratives of individual sessions — what the visitor did, where they stalled, how it ended — with classified intent, outcome and friction points.

Needs Growth or higher for data

When to use it

For qualitative colour once the quantitative tools have localised a problem. Filter outcome: 'abandoned_cart' to read what abandoners experienced, or gatedBy: 'high_frustration' for the angriest sessions.

Ask for it like this

You do not name the tool. These are questions that make an assistant reach for it:

What did people who abandoned their cart actually experience?
Summarise my most frustrated sessions.

How to read the result

  • topFrictionPoints rolls friction locations up across the window, so "what are visitors struggling with this week" is one call rather than twenty.
  • The classified fields make individual sessions aggregable, which raw journeys are not.

What will make you wrong

Caveats

  • Coverage is deliberately gated. Only converted, high-frustration or high-engagement sessions are summarised, capped per day and generated nightly. The absence of a summary says nothing about a session.
  • Summaries are an overlay. The structured events remain ground truth — verify a surprising claim with get_session_journeys before acting on it.

Parameters

outcome
enum
Filter by the classified outcome.
convertedabandoned_cartbouncedleft_no_actionother
intent
enum
Filter by the classified intent.
browsecomparepurchasesupportunknown
gatedBy
enum
Filter by why the session qualified for summarisation.
convertedhigh_frustrationhigh_engagement
limit
numberdefault: 20
Maximum summaries to return. Cap 100.

Shared parameters it accepts

Documented in full on the tool reference index.

Date window — defaults to the last 30 days.

dateFrom
stringdefault: 30 days ago
Start of the window, YYYY-MM-DD. Inclusive.
dateTo
stringdefault: today
End of the window, YYYY-MM-DD. Inclusive.

Every tool also accepts merchantId, which resolves from your connection. You never pass it.

What comes back

JSON. These are the top-level keys; date-scoped tools additionally carry a _meta block with the server date and the data coverage envelope.

summaries
Per session: the narrative, plus classified intent, outcome, frictionPoints and keyProducts.
topFrictionPoints
Friction locations rolled up across the whole window.
note
That summaries are generated over stored events, which remain the ground truth.

Pairs with

get_session_journeys