Your tracking console, queried in English.

TrackLayer ships with an MCP server (32 tools) so Claude, Cursor, and any MCP-compatible agent can read your events, write replay mutations, and debug drops end-to-end. Bring your own model — we don't charge per token.

§ 01

An incident, in chat.

CLAUDE SONNET 4.6 · MCP · 32 TOOLS LOADED

CHAT · #growth-ops

glass-house.co · anna@glass-house.co

USER

Why did Meta revenue drop 40% yesterday?

TOOL

⟶ brain.query_anomaly

{"platform":"meta","metric":"revenue","window":"24h","threshold":"3σ"}

tool call · 48ms

TOOL

⟶ supabase.select

SELECT platform, SUM(value) FROM events
WHERE platform='meta'
  AND received_at > now() - interval '48h'
GROUP BY 1, DATE(received_at)

query · 142ms · 18 rows

CLAUDE

Meta CAPI revenue dropped 42% from $18,420 to $10,671 between 14:00–18:00 UTC yesterday. Three things moved together:

CLAUDE

Purchase event volume fell 38% — but only on Meta, not GA4. Pixel is fine; CAPI ingestion lagged.
P95 latency to Meta spiked from 118ms to 2.4s over the same window.
Shopify deployed a new theme at 13:58 that broke the checkout page's `data-product-id` attribute — every purchase missed `content_ids`, so Meta's match quality fell below threshold for attribution.

CLAUDE

→ Rollback the theme deploy (checked.io/glass-house/deploy-0412).
→ Or patch the attribute selector and I can re-send the 840 affected events with `content_ids` populated.

USER

Re-send them.

TOOL

⟶ tracklayer.replay

{"merchant":"glass-house","platform":"meta","event_ids":[…840],"mode":"enrich_and_resend"}

mutation · 8.2s · 831/840 delivered

CLAUDE

831 events resent successfully. 9 failed with 4xx (probably already-matched duplicates — Meta's dedup caught them). Revenue should reconcile in the next batch. Anything else?

Ask anything about your pipeline…

⌘⏎

§ 02