How Aggregation Works

MCPHub pre-aggregates analytics data into hourly snapshots to keep dashboard queries fast, even with millions of raw tool call records.

The aggregation pipeline

tool_calls (raw)
       ↓
analytics_aggregator agent
       ↓
analytics_snapshots (hourly, per-server)
       ↓
Redis cache (5 min TTL)
       ↓
Analytics API → Dashboard

Hourly snapshots

The analytics_snapshots table stores one row per server per hour, containing:

• call_count — total tool calls in that hour
• error_count — tool calls with status = error
• avg_latency_ms — mean duration
• p95_latency_ms — 95th percentile duration
• Top 10 tools and their call counts

This means analytics queries are aggregation over at most ~720 rows (30 days × 24 hours) per server, rather than scanning millions of raw tool call rows.

When aggregation runs

The analytics aggregator runs:

• Daily at 02:00 UTC via Vercel Cron — processes the last 24 hours of tool calls for all workspaces
• On-demand via POST /api/v1/admin/evaluate-alerts (requires CRON_SECRET bearer token)

This means analytics data is at most ~24 hours behind for the current day. Yesterday's data is fully aggregated.

Redis cache

Analytics API responses are cached in Redis with a 5-minute TTL. This means:

• Repeated dashboard loads hit the cache, not the database
• Cache is scoped per workspace and per query parameters
• You may see data up to 5 minutes old during peak usage

Raw data access

For real-time or fine-grained analysis, use the tool call audit log directly. It contains every individual tool call with full detail.