New Relic runs a hosted MCP server at mcp.newrelic.com. Its centre of gravity is NRQL — the query language — which means an agent can go from a vague question to a precise timeseries without you remembering the syntax.
https://mcp.newrelic.com/mcp
Claude Sonnet 4.5
Chat with 60+ AI models on the same workflow — switch to a different model mid-conversation and re-run the same prompt, or use Compare mode to put several side-by-side and balance quality vs. cost.
A New Relic User API key, sent as an api-key header. The key carries your account scope, so it decides which accounts the agent can read.
How models use it and what it is built for.
The server is built around New Relic’s query layer rather than around dashboards. NRQL is the primary tool: an agent composes a query against events, metrics, logs or traces and reads the result, which covers most of what the platform can answer. Entity search sits alongside it, so a model can resolve "the checkout service" to a real entity GUID before querying it — the step that otherwise produces confidently wrong answers. Alert and incident state is readable for triage, and APM data covers throughput, error rate and transaction traces. Because NRQL spans every telemetry type in one syntax, this server tends to need fewer tool round-trips than observability servers that split logs and metrics into separate tools, which is a real advantage on token cost.
Typical tools an AI model can call. Exact names vary by version.
Copy any of these into MCP Agent Studio after connecting.
Write and run the NRQL for error rate by transaction on the checkout service, last 24 hours.
Which entities in this account have an open alert right now?
What is the slowest transaction in the API service this week, and how has it trended?
Compare Apdex for the web tier before and after yesterday’s deploy.
This is not a single-model product: you get the same MCP connection with 60+ models (Claude, GPT, Gemini, DeepSeek, open-weight, and more), you can switch mid-conversation, and you can open Compare mode to run the same prompt against multiple models at once. The card above is a suggested starting point for this server — not the only choice.
Default pick for New Relic
Claude Sonnet 4.5
NRQL has its own FACET and TIMESERIES semantics that superficially resemble SQL. Sonnet 4.5 writes it correctly; weaker models produce SQL-shaped queries that fail or silently mislead.
Listing tools proves the server is reachable, not that a model can work with it. Evals go further: they read every tool on the server, write a test suite from its real schemas, and run it — code decides pass/fail on the responses (schema conformance, error codes, pagination, result caps) while a scoring model grades plain-English tasks driven through the tools.
Get a pass/fail report per tool with the evidence behind each verdict — and replay the same suite after every schema change. Destructive tools are excluded from the run.
Run evalsOpen MCP Agent Studio with the connection pre-filled. Add your token, pick any of 60+ models, and start chatting — no install required.
Open Agent StudioCommon questions about connecting, scoping and using it safely.
New Relic’s hosted MCP server at mcp.newrelic.com. It gives an AI assistant NRQL query access plus entity search, APM metrics and alert state, so you can investigate telemetry in conversation rather than in the query builder.
No — that is largely the point. You describe what you want and the model writes the NRQL. It is still worth reading the query it produces, because a subtly wrong FACET or time window returns a plausible number rather than an error.
A User API key, passed as an api-key header. It inherits that user’s account access, so create a dedicated key with the narrowest account scope you need rather than reusing an admin key.
New Relic routes almost everything through NRQL, so one flexible tool answers most questions. Datadog exposes more separate, purpose-built tools. In practice New Relic needs fewer round-trips, Datadog needs less query correctness from the model.
The server is oriented toward reading telemetry and alert state rather than mutating configuration. Treat the key as the real boundary: scope it to read-only access and the question stops being about what the tools might allow.
Datadog
Query metrics, logs, traces and monitors from a conversation instead of a dashboard.
Honeycomb
Query high-cardinality traces and let an agent chase the outlier, not the average.
Grafana
Query Prometheus and Loki, search dashboards, triage alerts in chat.
Sentry
Triage issues, stack traces and releases from chat without opening ten browser tabs.