What to bring in
The signals that matter most when agents are investigating or writing RCAs:- NRQL queries via NerdGraph for any metric, event, or log New Relic holds.
- APM service health (throughput, error rate, apdex) for the services involved in an incident.
- Alerts (policies, conditions, incidents) to surface firing alerts as insight candidates.
- Entity relationships for service and host topology.
Which path to pick
Start with REST
Point at the NerdGraph endpoint with an OpenAPI wrapper, or use the REST v2 API directly.
Move to MCP when you need more
Wrap NRQL behind named tools (for example
new_relic_service_latency) so agents never see raw query strings.Quick recipe: REST path
Create a New Relic user key
In New Relic, create a read-scoped user key. No admin or license key is required.
Register the endpoint
In the RubixKube console, go to Integrations → Custom REST and register NerdGraph (
/graphql) or the REST v2 API.Enable the calls you care about
Start with NRQL execution, alert state, and entity search. Leave everything else disabled.
Example tools worth having
nr_service_health(service, window)— wraps an NRQL query for throughput, error rate, and apdex.nr_top_errors(service, window)— returns the most frequent error fingerprints for the window.nr_firing_incidents(service)— returns incidents currently firing on a service.nr_entity_relationships(service)— returns upstream and downstream entities for topology context.
Related guides
Custom REST Integrations
The full REST flow with auth options and approval rules.
Custom MCP Servers
When you want richer tool schemas and multi-call workflows.
Skills
How to compose New Relic queries into custom workflows.