New Relic MCP Server

Install

• One‑liner: npx -y newrelic-mcp

Server config (drop-in for most MCP clients)

{
  "mcpServers": {
    "newrelic": {
      "command": "npx",
      "args": ["-y", "newrelic-mcp"],
      "env": {
        "NEW_RELIC_API_KEY": "YOUR_API_KEY",
        "NEW_RELIC_ACCOUNT_ID": "YOUR_ACCOUNT_ID"
      }
    }
  }
}

Environment variables

• NEW_RELIC_API_KEY: New Relic User API Key (required)
• NEW_RELIC_ACCOUNT_ID: Default account ID (optional; can be provided per tool call)

Tools (copy for MCP.so "Tools" section)

• run_nrql_query: Execute NRQL against an account.
• run_nerdgraph_query: Run arbitrary NerdGraph GraphQL.
• search_entities: Search across entities (type, tags, account).
• **get_entity_details: **Deep detail for any entity GUID (relationships, golden metrics, alert context).
• list_apm_applications: List APM apps via GraphQL.
• list_synthetics_monitors: List Synthetics monitors (filter by type).
• create_browser_monitor: Create a Synthetics Browser monitor.
• list_alert_policies: Enumerate alert policies.
• list_open_incidents: Fetch open incidents (filter by priority).
• acknowledge_incident: Acknowledge a specific incident.

REST v2 (region-aware, auto‑pagination)

• create_deployment
• list_deployments_rest
• delete_deployment (requires confirm: true)
• list_apm_applications_rest
• list_metric_names_for_host
• get_metric_data_for_host
• list_application_hosts
• get_account_details: Convenience account info

Recommended System Prompt:

You are an SRE copilot using New Relic via MCP tools. Follow these rules:
1) Choose the right interface:
   - Use run_nrql_query for aggregate analytics and timeseries.
   - Use run_nerdgraph_query for graph-shaped data and entities/alerts APIs.
   - Prefer REST v2 tools for deployments and legacy APM metrics; set region as needed.
2) Always include target_account_id unless a default is configured.
3) For large result sets, enable auto_paginate and surface page info.
4) Be safe: require explicit user confirmation for delete_deployment (confirm=true).
5) For incidents:
   - Use list_open_incidents to triage by priority.
   - Use acknowledge_incident only after user approval.
6) For synthetics:
   - Use create_browser_monitor with name, url, frequency, locations.
7) Return concise summaries with a short “What I did” and “What next” suggestion.

Example Prompts

• “Run NRQL to get requests and error rate for app ‘web’ over the last 30 minutes; include a short analysis.”
• “List open incidents with priority CRITICAL and HIGH; summarize by policy.”
• “Acknowledge incident 1234 with comment ‘false positive after deploy rollback’.”
• “Create a Synthetics Browser monitor named ‘Homepage SLA’ for https://example.com every 5 mins from US_EAST_1 and EU_WEST_1.”
• “Record deployment for application_id 987654 with revision 9f3c2ab; include changelog and user.”
• “List APM applications via REST; filter language=nodejs; auto-paginate across regions US then EU.”
• “For app 987654 host 321, list metric names, then fetch CPU and memory timeslices in the last 30 minutes.”

Testing

• Tested to 90%+ Code Coverage
• Behavior defined in Gherkin; evaluated with AI evals testing.
• Tool discovery and input schema validation covered. Pagination, error handling, and performance scenarios included.