generated: '2026-08-13'
method: searched
source: https://apstal.com/docs/mcp
note: >-
  There is no OpenAPI to derive an entity graph from. Apstal's data model is a single wide
  analytics events table queried with read-only SQL through the MCP `execute_sql` tool; the
  documentation names the key columns and says the full schema is 60+ columns, retrievable at
  runtime with the `get_schema` tool. Only the columns the provider actually names are listed
  below - the remaining columns are recorded as unknown rather than invented.
entities:
- name: events
  kind: table
  store: BigQuery
  description: >-
    One row per tracked interaction. Every query is automatically scoped to the caller's
    project_id, filtered to exclude bot traffic, and restricted to SELECT.
  primary_scope: project_id
  columns_documented:
  - name: event_type
    type: string
    values: [pageview, click, scroll, exit, purchase, rage_click]
    note: The docs list these with "etc." - the enumeration is not exhaustive.
  - name: visitor_id
    type: string
    role: groups sessions to an anonymous visitor
  - name: session_id
    type: string
    role: groups events into a session; used for user-path analysis
  - name: page_url
    type: string
  - name: page_path
    type: string
  - name: page_title
    type: string
  - name: country
    type: string
  - name: city
    type: string
  - name: device_type
    type: string
  - name: browser
    type: string
  - name: os
    type: string
  - name: fcp
    type: number
    role: Core Web Vitals - First Contentful Paint
  - name: lcp
    type: number
    role: Core Web Vitals - Largest Contentful Paint
  - name: cls
    type: number
    role: Core Web Vitals - Cumulative Layout Shift
  - name: ttfb
    type: number
    role: Core Web Vitals - Time To First Byte
  - name: utm_source
    type: string
  - name: utm_medium
    type: string
  - name: utm_campaign
    type: string
  - name: amount
    type: number
    role: ecommerce revenue
  - name: currency
    type: string
  - name: metadata
    type: json
    access: JSON_EXTRACT_SCALAR
    role: additional per-event fields
  columns_total: 60+
  columns_undocumented: true
  runtime_schema_tool: get_schema
relationships:
- from: project
  to: events
  kind: has_many
  via: project_id
- from: visitor
  to: session
  kind: has_many
  via: visitor_id
- from: session
  to: events
  kind: has_many
  via: session_id
concepts:
- name: project
  description: >-
    A tracked domain. Created via the MCP `register_domain` tool or the dashboard; identified by
    a UUID Project ID that is embedded in the tracking snippet as data-project.
  id_shape: UUID
  source: https://apstal.com/docs/mcp
- name: session replay
  description: >-
    rrweb-format chunks uploaded to POST /api/v1/stream and stored at
    session-replays/<projectId>/<sessionId>/chunk_<n>.json.gz. Retained 30 days by default,
    then deleted by a cleanup job.
  source: https://apstal.com/docs/api
- name: ai_agent tag
  description: >-
    Sessions produced by agentic browsers are tagged ai_agent with the agent name. Available
    from the tracker version released 2026-08-07; earlier sessions carry no tag.
  detected_agents: [ChatGPT Atlas, Perplexity Comet, Claude agent, Manus, Cursor, Browser Use, Skyvern, Arc, Fellou, Genspark, Sigma]
  source: https://apstal.com/docs/ai-agents
access_rules:
- SELECT only; no DML is accepted by execute_sql.
- project_id is injected server-side when the caller omits it.
- Bot traffic is filtered out of query results by default.
- A date filter is applied by the server.
source_note: >-
  The SQL-safety bullets on /docs/mcp render as untranslated i18n keys
  (DocsMcp.sqlSafety.selectOnly, .isolation, .botFilter, .dateFilter), so the four rules above
  are read from their key names plus the surrounding prose rather than from published sentences.