AI导哈

可观测性数据查询MCP服务器

New Relic MCP服务器通过标准MCP协议让AI助手以自然语言查询New Relic可观测性数据，支持指标查询、错误分析、性能追踪和告警检查等开发运维场景。

  • by atamano
  • TypeScript
  • 本地部署
  • 更新于
在 GitHub 查看 atamano/newrelic-mcp

使用说明

New Relic MCP Server

An MCP server that gives AI assistants access to your New Relic observability data. Query metrics, investigate errors, analyze performance, and check alerts — all through natural language.

Tools

Tool Description
query_nrql Run any arbitrary NRQL query
get_app_health Health overview: throughput, error rate, response time, Apdex
get_errors Recent errors/exceptions with class, message, and count
get_error_details Detailed errors with stack traces, filterable by class or transaction
get_slow_transactions Slowest transactions with avg/max/p95 duration
get_throughput Request throughput (rpm) and error rate over time
get_database_performance Transactions with the highest database time
get_http_status_breakdown HTTP status code distribution
get_alerts Recent alert incidents with title, priority, and state

Vercel apps (log drain)

For Vercel projects whose runtime logs reach New Relic through a Vercel log drain to the Log API. They have no APM, so the tools above do not see them; these read the Log rows the drain writes (projectName, requestId, proxy.*, and Vercel's REPORT … Duration: N ms line).

Tool Description
get_vercel_health Per project: requests, 5xx count and rate, cache hit share, function invocations and p50/p95 duration
get_vercel_errors Routes that answered 5xx, and the error/warning messages functions logged
get_vercel_slow_routes Routes ranked by total function time, with calls, average and p95 duration

Prerequisites

  • Node.js 18+
  • A New Relic account with a User API key (starts with NRAK-)

Setup

git clone https://github.com/indiefoundry/newrelic-mcp.git
cd newrelic-mcp
pnpm install
pnpm run build

Configuration

The server reads configuration from environment variables. See .env.example for reference.

Note: The server does not load .env files automatically. Environment variables must be set in your shell, passed by the MCP client, or exported before running.

Variable Required Description
NEW_RELIC_API_KEY Yes User API key (starts with NRAK-)
NEW_RELIC_ACCOUNT_ID Yes Your numeric account ID
NEW_RELIC_REGION No US or EU (default: EU)

Usage

Claude Desktop

Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json on macOS):

{
  "mcpServers": {
    "newrelic": {
      "command": "node",
      "args": ["/absolute/path/to/newrelic-mcp/dist/index.js"],
      "env": {
        "NEW_RELIC_API_KEY": "NRAK-...",
        "NEW_RELIC_ACCOUNT_ID": "1234567",
        "NEW_RELIC_REGION": "EU"
      }
    }
  }
}

Claude Code

claude mcp add newrelic \
  -e NEW_RELIC_API_KEY=NRAK-... \
  -e NEW_RELIC_ACCOUNT_ID=1234567 \
  -e NEW_RELIC_REGION=EU \
  -- node /absolute/path/to/newrelic-mcp/dist/index.js

Other MCP Clients

This server uses stdio transport. Any MCP-compatible client can connect by spawning the process:

NEW_RELIC_API_KEY=NRAK-... NEW_RELIC_ACCOUNT_ID=1234567 node dist/index.js

Development

pnpm run dev        # run with tsx in watch mode (auto-restarts on changes)
pnpm run build      # compile TypeScript
pnpm run start      # run compiled output
pnpm run typecheck  # type-check without emitting

Example Queries

Once connected, you can ask your AI assistant things like:

  • "What errors happened in the last hour?"
  • "Show me the slowest API endpoints for my-app"
  • "What's the current error rate and throughput?"
  • "Are there any active alerts?"
  • "Run this NRQL query: SELECT count(*) FROM Transaction FACET appName SINCE 1 day ago"

Architecture

The server consists of two modules:

  • src/newrelic-client.ts — Wraps the NerdGraph API (New Relic's GraphQL API). Handles NRQL query execution, response validation, GraphQL error surfacing, and request timeouts (30s).
  • src/index.ts — MCP server that registers all tools with validated input schemas and connects via stdio transport.

License

MIT

相关 MCP