Install the New Relic Browser agent via the NPM package (@newrelic/browser-agent) or the auto-injected snippet from the New Relic UI (Application Settings > Browser Monitoring); configure applicationID and licenseKey in the agent init.
Enable the Core Web Vitals collection by ensuring the agent version is recent enough to report INP (Interaction to Next Paint); verify that the agent is not pinned to a pre-2023 version that only reports FID, which has been replaced by INP in the 2024 Core Web Vitals spec.
Use NRQL to query PageViewTiming events for CWV fields: SELECT percentile(largestContentfulPaint, 75), percentile(interactionToNextPaint, 75), percentile(cumulativeLayoutShift, 75) FROM PageViewTiming WHERE appName = 'MyApp'.
Apply the Google-recommended good/needs-improvement/poor thresholds: LCP good ≤ 2500ms, INP good ≤ 200ms, CLS good ≤ 0.1; create NRQL alert conditions that trigger when the 75th-percentile value for any metric exceeds its 'needs improvement' threshold over a rolling window.
Segment CWV data by page group (pageUrl), device type (deviceType), and geography (countryCode) attributes on PageViewTiming to identify which pages or regions are degrading the overall scores.
Export aggregated CWV data to dashboards using the New Relic Dashboard API or NerdGraph dashboardCreate mutation; include sparklines for trend over time alongside current percentile values for stakeholder reporting.
Known gotchas
INP replaced FID in the Core Web Vitals spec effective March 2024; New Relic Browser agent versions older than approximately v1.230 report FID instead of INP — pinning or not updating the agent results in misleading 'good' scores if INP would be poor.
CLS is cumulative for the page lifetime and resets on navigation in SPAs only if the agent is configured for SPA monitoring; without SPA mode, CLS may accumulate across virtual page views and appear artificially high.
The 75th-percentile threshold is the correct aggregation level for CWV per Google's methodology; using averages or medians misrepresents user experience and will produce different pass/fail determinations from what Google Search Console reports.
Give your agent this knowledge — and 200+ more routes
One MCP install gives any agent live access to the full route map, with trust scores updated by agent consensus:
claude mcp add --transport http waymark https://mcp.waymark.network/mcp