Custom New Relic MCP Server
DevOps Engineer @ Accuris
Built a Model Context Protocol server for New Relic that beat the existing integration on query accuracy, token cost, and response latency.
Impact
- Higher NRQL query-generation accuracy than the existing integration.
- Lower token consumption per request, reducing running cost.
- Lower response latency for day-to-day observability questions.
The problem
The existing New Relic integration our engineers reached for was expensive and imprecise. It generated NRQL that frequently missed the intent of the question, burned through tokens on verbose round-trips, and was slow enough that people went back to clicking through the UI instead of asking for what they needed.
The approach
I built a custom Model Context Protocol (MCP) server for New Relic from the ground up, tuned specifically for how our teams actually query observability data. It exposed focused, well-described tools, constrained the model toward valid NRQL, and trimmed the context sent on each call so responses came back faster and cheaper.
Because it spoke MCP, it dropped straight into the same client surface as the rest of our tooling — and later became one of the servers unified by the MCP mesh.
The impact
- More accurate NRQL generation than the solution it replaced.
- Fewer tokens per request — a direct, recurring cost saving.
- Lower latency, which is what actually got people to adopt it.