
Preflight
io.github.newrelic-experimental由 GitHub 账号 newrelic-experimental 发布v1.71.0更新于 Nov 4, 2026
New Relic MCP server for observing AI coding assistants (Claude Code, Cursor, Copilot, etc.)
发布方自己的服务:Preflight
概览
New Relic 推出的本地 MCP 服务器,用于采集并分析 AI 编程助手的工具调用、成本与效率,可选接入 New Relic 云端。
- 可以做什么
- 通过 nr_observe_* 工具查询已采集的会话遥测:操作、工具调用、成本与效率数据
- 检查安装健康状态,并在明确允许时安装用于采集工具调用的钩子
- 在 localhost:7777 运行本地实时仪表板,展示会话、花费与效率评分
- 配置 New Relic 账号后可部署预置仪表板与告警
- 提供成本明细、反模式检测、效率评分与每周辅导报告
- 适用场景
- 适合使用 Claude Code、Cursor、Copilot 等 AI 编程助手、并希望了解其做了什么、花了多少钱、在哪里浪费时间的人。本地模式无需账号;只有需要团队汇总、告警和跨会话历史时才接入 New Relic。
- 运行要求
- 通过 npx 从 npm 包 @newrelic/preflight 在本地运行;需要 Node.js v22 或更高版本以及受支持的 AI 编程工具。本地模式无需账号。cloud 或 both 模式需要 NEW_RELIC_ACCOUNT_ID 与 NEW_RELIC_LICENSE_KEY,并用 NR_AI_MODE 指定遥测去向。部署仪表板和告警还需单独的 New Relic 用户 API 密钥。
安装
在 SourceWeft 中
- 打开 控制台中的 Preflight,将其添加到工作区。
- 为需要使用其工具的对话启用该服务。
Desktop only,通过 STDIO。 STDIO 服务会启动本地进程,因此需要 SourceWeft 桌面宿主。
其他 MCP 客户端
参照 仓库 中的启动说明。
README
Preflight
Observability for AI Coding Assistants
[License] [Node 22+] [Local First] [Dashboards Included]
Docs • What's New • Examples • Community • Contributing
Why Your AI Tool Needs Observability
Your AI coding assistant makes hundreds of decisions every session — what to read, what to edit, when to run commands. But you can't see any of it. You know it was fast, but was it efficient? You got a PR merged, but how much did it cost? You fixed a bug, but did it get stuck in a loop first?
Preflight is observability for agentic coding — the actions, cost, and efficiency of your AI coding assistant as it works. See exactly what's happening, how much it costs, and where your AI is wasting time.
Local-first by design. Preflight runs entirely on your machine and sends your data nowhere by default. A live dashboard at localhost:7777 shows your sessions in real time, fully offline. Connect a New Relic account only when you want more — team rollups, alerting, and cross-session history. You choose: local-only, New Relic, or both.
Demo
[Preflight dashboard animation]
See cost breakdown, efficiency scoring, anti-patterns, and live session tracking in action.
What You Get
Visibility
- Every action captured — file reads, edits, commands, searches
- Live session dashboard — see what's happening right now
- Historical trends — analyze patterns over weeks and months
Cost Control
- USD spend tracking — per session, day, and week
- Per-model and cache breakdown — know which models cost most and how efficiently context is reused
- Budget alerts — get notified before you overspend
- Forecasting — project monthly burn rate
Efficiency Insights
- Efficiency score — 0–100 score per task, based on how directly the AI worked
- Anti-pattern detection — catches re-reads, blind edits, stuck loops
- Personalized recommendations — optimize your AI workflow
- Weekly coaching reports — narrative analysis vs. your historical baseline
Dashboards
- Local dashboard — live session view at
localhost:7777, no account required - 8 pre-built New Relic dashboards — deploy in seconds (New Relic mode):
- Overview — session stats, cost summary, top tools
- Personal — 30-day self-reflection scoped to you
- Session Detail — deep-dive into a single session's tool calls
- Team View — aggregated cost and efficiency across developers
- Manager View — high-level team metrics, no tool-call content
- Adoption & Cost — adoption momentum, spend, skill and MCP usage, and per-developer outcomes for engineering managers
- Platform Comparison — Claude Code vs. Cursor vs. Windsurf, etc.
- Security Audit — audit trail of sensitive file access
Quick Start
1. Install
Using Smithery? Installing Preflight from the Smithery MCP registry wires up the MCP server for you, but Smithery has no mechanism to write Claude Code hooks. After install, either run
preflight installin a terminal, or set"allowHeadlessInstall": truein your Preflight config (orNEW_RELIC_AI_ALLOW_HEADLESS_INSTALL=true) and ask Claude Code to call thenr_observe_install_hooksMCP tool. The tool refuses by default, because any connected MCP client can call it. With the flag on, it offers to do this automatically oncenr_observe_healthreportssetup_required: true. Then restart Claude Code to activate monitoring.
2. Run setup
The wizard defaults to local mode — press Enter through the prompts and you're set. It wires Preflight into your AI tool (hooks + MCP server) and writes config to ~/.newrelic-preflight/. Takes under a minute, no account required.
Using GitHub Copilot? The wizard also asks to install Copilot hooks — saying yes configures both the Copilot CLI and VS Code Copilot Chat automatically (hooks, MCP registration, and the fix for VS Code's hook double-capture), so Copilot gets the same tool-call and cost metrics Claude Code does. Run it standalone anytime with
preflight install --copilot. See docs/ADAPTERS.md for details.
When prompted, pick a mode:
3. Start coding
Restart your AI tool — hooks and the MCP server load at session start. Every tool call is captured automatically. Open http://localhost:7777 to watch your session live.
Using Kiro? Add Preflight as a Kiro Power for the
nr_observe_*query tools — see the doc for install steps and how to add automatic tool-call capture on top.
Using Claude Code? You can skip the npm install above and add Preflight as a Claude Code plugin instead:
Other ways to install
Cursor — Add to Cursor installs the MCP server via a one-click deeplink; run preflight setup afterwards for hook capture.
Homebrew (macOS):
Run preflight setup afterwards for hook capture.
Works With
Claude Code • Amazon Kiro • Amazon Q Developer CLI • Factory Droid • OpenAI Codex • opencode • Kilo Code • Pi • GitHub Copilot • GitHub Copilot SDK • GitHub Copilot app • Google Gemini CLI • Cursor • Windsurf • Google Antigravity • Zed • Continue.dev • Cline
Coverage isn't uniform — some platforms capture every built-in tool call, others (Zed, Continue.dev, Cline) only see calls routed to Preflight's own MCP tools. See ADAPTERS.md for what each platform can and can't observe, and per-platform setup steps.
Connect New Relic (optional)
Local mode is fully featured on its own. Connect a New Relic account to unlock:
- Team & manager dashboards across multiple developers
- Alerting on cost spikes, low efficiency, and stuck loops
- Cross-session history, trends, and weekly coaching reports
Re-run preflight setup and choose cloud or both, or configure it non-interactively:
EU accounts add --eu. FedRAMP accounts add --fedramp. Japan accounts add --jp.
A key passed with --license-key is visible to every local user through ps. For scripts, read it from a file instead (--license-key-file ~/.nr-key, which warns if other users can read the file) or set NEW_RELIC_LICENSE_KEY and NEW_RELIC_ACCOUNT_ID in the environment. The flag wins over the file, and the file wins over the environment. The environment is used only when config.json has no stored licenseKey or licenseKeyCommand, so a shell that already exports these variables for an APM agent never overwrites your stored key. To replace a stored key, pass --license-key-file. With --copilot, the key is still handed to copilot mcp add as an argument, so it is briefly visible in ps for that one call.
Then deploy the prebuilt dashboards:
You'll need a license key (telemetry ingest) and your account ID, plus a user API key (NRAK-…) to deploy dashboards and alerts. See ADVANCED.md for alerts, OTLP export to other backends, and Terraform.
No dashboard until you run
deploy-dashboards. Cloud mode ships telemetry to New Relic as soon as it's configured, but nothing creates a dashboard automatically — that's the separate step above, and it needs a different credential (a user API key, not your license key). Until you run it, there's no UI to look at. In the meantime, query the raw events directly in New Relic's Query Builder, e.g.:Once
deploy-dashboardssucceeds, find the dashboards under your New Relic account's Dashboards section.
Data ingest note: Telemetry sent to New Relic counts against your account's data ingest. On paid plans, standard ingest rates apply. Monitor your usage under NR One → Data Management → Data Ingestion.
Requirements
Required
- Node.js v22 or higher (get it or use nvm)
- An AI coding tool (Claude Code recommended for deepest integration)
Optional
- New Relic account — only for
cloud/bothmode. Skip it to run local-only (the default). - User API key (
NRAK-…) — only needed to deploy dashboards and alerts
Other Commands
Add --project to install/uninstall to scope changes to the current directory only.
WSL users: preflight setup will ask which Claude Code you're running. You can also set it explicitly:
--windows-cc— Windows Claude Code (the desktop app); useswsl.exehooks and Windows paths--linux-cc— Linux Claude Code installed via npm inside WSL
Documentation
- ADVANCED.md — Configuration, dashboards, alerts, Terraform
- SCORECARDS.md — New Relic Scorecard rules for team attribution
- ARCHITECTURE.md — Data flow, component reference, and operating modes
- ADAPTERS.md — Per-platform integration mechanism, setup steps, and known gaps
- FLEET_DEPLOYMENT.md — Roll out the Claude Code plugin to many machines with managed settings and MDM
- TROUBLESHOOTING.md — Common setup and connection problems, and how to fix them
- CONTRIBUTING.md — Development, testing, submitting PRs
- SECURITY.md — Security guidelines and best practices
- PRIVACY.md — Data collection inventory and pre-cloud checklist
From Source
Develop, test, or run the latest unreleased version:
Then run preflight setup as usual.
License
Preflight is open source under the Apache License 2.0.
Contributing
We welcome contributions! See CONTRIBUTING.md for how to get started. Join the New Relic Community to share ideas, ask questions, or discuss features.
Built by New Relic • Designed for developers who use AI
来源:README.md,提交 861a9a7
工具
0版本历史
22- v1.71.0最新Nov 4, 2026
- v1.70.0Nov 3, 2026
- v1.68.0Oct 29, 2026
- v1.67.3Oct 26, 2026
- v1.67.2Oct 14, 2026
- v1.67.1Oct 13, 2026
- v1.67.0Oct 12, 2026
- v1.65.0Oct 12, 2026
- v1.63.4Oct 9, 2026
- v1.63.1Oct 8, 2026
- v1.61.0Oct 8, 2026
- v1.60.0Oct 8, 2026
- v1.59.4Oct 7, 2026
- v1.59.0Oct 7, 2026
- v1.58.3Oct 6, 2026
- v1.57.3Oct 6, 2026
- v1.57.0Sep 18, 2026
- v1.56.0Sep 18, 2026
- v1.55.1Sep 17, 2026
- v1.55.0Sep 17, 2026
- v1.54.0Sep 16, 2026
- v1.53.1Sep 16, 2026
评分
已满足适用检查项的 50%
该评分衡量此收录条目的完整程度,依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价,也不表示服务器安全或当前可用:可连通性是带日期的检查结果,评分也不包含安装量或用户评价。
- 近 30 天内可连通25 分
请保持端点能够响应 MCP 握手;SourceWeft 每 30 天探测一次。
- 已声明认证方式10 分
请在注册中心条目中声明凭据请求头,或保持端点能够响应探测,以便判定其认证方式。
- 已列出工具15 分
请让服务器在无凭据时响应 tools/list,或发布可在沙箱中无凭据启动的软件包。
- 描述必需15 分
- 图标10 分
- 文档15 分
- 许可证10 分
其他 Preflight 集成
Preflight
social.preflight官方发布者
托管式社交媒体排程服务器,让助手起草、按规则校验、排程并向 12 个平台发布帖子。

SQLite rebuild copy-map preflight
由 GitHub 账号 showflexinc 发布
远程预检服务,在重建执行前检查 SQLite 表重建复制映射的结构。

Rust MCP Preflight
由 GitHub 账号 Wolfe-Jam 发布
在发布前检查 MCP Registry 条目:README 中的 mcp-name 行、server.json 的发布规则以及服务器名称。

Multipart row preflight
由 GitHub 账号 showflexinc 发布
CSV row-structure inspection (inspect_parts) and manifest comparison (compare_manifest).

Tool-dialogue preflight
由 GitHub 账号 showflexinc 发布
针对JSONL工具对话训练数据的结构化预检。AI 翻译