de en es ja ko ru zh

MCP Infra Lens

Name: MCP Infra Lens
Author: oaslananka

by oaslananka

Overview Schema Source Related Servers Score Discussions Admin

TypeScript

Hybrid

mcp-infra-lens

通过支持基准的 MCP 工具解释 SSH Linux 事件。

npm version npm downloads License: MIT Node 20+ MCP SDK Azure Pipelines

演示

mcp-infra-lens demo

当 Claude 询问“prod-01 有什么问题？”时，analyze_server 的示例响应：

{
  "host": "prod-01.internal",
  "health_score": 42,
  "summary": "Found 2 anomalies on prod-01.internal. Most urgent signal: CPU is at 91% (3.4σ above baseline 28.2%). Load is 7.2/6.8/5.1. Top CPU consumer: java (87%).",
  "anomalies": [
    {
      "metric": "cpu",
      "severity": "high",
      "value": 91,
      "z_score": 3.4,
      "explanation": "CPU is at 91% (3.4σ above baseline 28.2%). Load is 7.2/6.8/5.1. Top CPU consumer: java (87%).",
      "recommendation": "Investigate java (PID 18432) and review application logs or scale-out options."
    },
    {
      "metric": "disk:/",
      "severity": "high",
      "value": 91,
      "explanation": "Disk / is 91% full (182GB/200GB).",
      "recommendation": "Run du -sh //* | sort -rh | head -20 and clean logs or temporary files."
    }
  ]
}

功能介绍

mcp-infra-lens 通过 SSH 连接到 Linux 主机，捕获实时基础设施快照，将其与最近记录的基准进行比较，并用通俗易懂的语言解释异常情况。

在不修改目标主机的情况下收集 CPU、内存、磁盘、网络、进程和操作系统数据
在 SQLite 中记录本地指标历史，用于基准测试、比较和趋势查询
在存在足够的基准样本后，使用 z-score 分析进行 CPU 异常检测
解释压力产生的可能原因，而不仅仅是原始指标值
支持通过 stdio 和可流式传输的 HTTP 进行 MCP 通信

工作原理

flowchart TD
    A["Claude / Cursor / VS Code / Windsurf"] --> B["mcp-infra-lens"]
    B --> C["server-core.ts"]
    C --> D["collector.ts"]
    C --> E["analyzer.ts"]
    C --> F["baseline.ts + db.ts"]
    D --> G["ssh.ts"]
    G --> H["Linux host over SSH"]
    F --> I["SQLite history + baselines"]

analyze_server 现在会在请求的 duration_minutes 内执行实际的采样收集，计算收集窗口内的 CPU 和内存压力平均值，持久化生成的快照，然后针对选定的基准运行异常检测。

工具

工具	功能	关键参数
`analyze_server`	收集采样快照、存储并解释异常	`connection`, `duration_minutes`, `include_processes`, `include_network`
`snapshot`	捕获并存储当前时间点的指标，不进行分析	`connection`
`record_baseline`	保存标记的健康状态样本以供将来比较	`connection`, `label`
`compare_to_baseline`	将当前状态与命名基准进行比较并解释差异	`connection`, `baseline_label`
`get_history`	从 SQLite 返回历史 CPU、内存或负载点	`host`, `metric`, `hours`, `label?`

快速入门

1. 通过 `npx` 运行

npx -y mcp-infra-lens

如果您固定在 1.0.1 版本，请升级到 1.0.2 或更高版本，以避免 Node 24 原生安装失败：

npx -y mcp-infra-lens@latest

2. Claude Desktop

已发布包：

{
  "mcpServers": {
    "infra-lens": {
      "command": "npx",
      "args": ["-y", "mcp-infra-lens"],
      "env": {
        "INFRA_LENS_DB": "/Users/you/.mcp-infra-lens/metrics.db"
      }
    }
  }
}

本地开发：

{
  "mcpServers": {
    "infra-lens": {
      "command": "node",
      "args": ["/absolute/path/to/mcp-infra-lens/dist/mcp.js"],
      "env": {
        "INFRA_LENS_DB": "/Users/you/.mcp-infra-lens/metrics.db"
      }
    }
  }
}

3. Docker

docker build -t mcp-infra-lens .
docker run --rm -it \
  -v "$HOME/.mcp-infra-lens:/home/appuser/.mcp-infra-lens" \
  mcp-infra-lens

配置

环境变量	默认值	描述
`INFRA_LENS_DB`	`~/.mcp-infra-lens/metrics.db`	SQLite 数据库路径。测试时使用 `:memory:`
`HOST`	`127.0.0.1`	HTTP 传输的绑定地址
`PORT`	`3000`	HTTP 传输的端口

健康评分

90-100: 健康，未检测到重大异常
70-89: 轻微或孤立的压力
40-69: 存在多个警告或正在进行重大问题
0-39: 严重状况，需要紧急补救

身份验证

SSH 输入模式支持：

密码验证
内联私钥验证
加密密钥的可选密码短语支持

凭据字段在写入 stderr 之前会从结构化日志中脱敏。

安全说明

SSH 收集在目标主机上是只读的
SSH 凭据永远不会存储在 SQLite 中
主机密钥验证在 v1 中是宽松的，以实现兼容性；生产部署应限制出站网络访问，并计划在后续版本中强制执行严格的主机验证
HTTP 传输没有内置身份验证；在任何非本地部署中，请绑定到回环地址并将其置于经过身份验证的反向代理之后

请参阅 SECURITY.md 了解报告策略和存储数据范围。

集成

Claude Desktop

{
  "mcpServers": {
    "infra-lens": {
      "command": "npx",
      "args": ["-y", "mcp-infra-lens"],
      "env": {
        "INFRA_LENS_DB": "/Users/you/.mcp-infra-lens/metrics.db"
      }
    }
  }
}

Cursor IDE

{
  "mcpServers": {
    "infra-lens": {
      "command": "npx",
      "args": ["-y", "mcp-infra-lens"]
    }
  }
}

VS Code (MCP 扩展)

{
  "inputs": [],
  "servers": {
    "infra-lens": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "mcp-infra-lens"]
    }
  }
}

Windsurf

{
  "mcpServers": {
    "infra-lens": {
      "command": "npx",
      "args": ["-y", "mcp-infra-lens"]
    }
  }
}

Docker (HTTP 传输)

docker run -d \
  -p 3000:3000 \
  -v $HOME/.mcp-infra-lens:/home/appuser/.mcp-infra-lens \
  ghcr.io/oaslananka/mcp-infra-lens:latest

然后配置您的 MCP 客户端以使用 http://localhost:3000。

Docker

捆绑的 Docker 镜像：

在单独的阶段构建 TypeScript 项目
在两个阶段中为容器架构重新构建 better-sqlite3
以非 root 用户 appuser 身份运行
将 SQLite 数据存储在 /home/appuser/.mcp-infra-lens/metrics.db 中

贡献

欢迎贡献。从 CONTRIBUTING.md 开始，然后使用：

docs/usage.md 获取工具示例
docs/architecture.md 获取组件映射
docs/testing.md 进行本地验证和发布检查
RELEASE _POLICY.md 获取 npm 和 MCP Registry 版本控制规则
AGENTS.md, CLAUDE.md, GEMINI.md, .github/copilot-instructions.md 和 .agent/rules/repository.md 获取特定于存储库的 AI 编码指导

运维 / CI 说明

azure-pipelines.yml 是规范的 CI 流水线，现在在 Node 20 和 Node 22 上运行 Quality 阶段，发布 JUnit 和 Cobertura 工件，并在 Node 20 上执行基于 Docker 的 SSH 端到端覆盖测试
.azure/pipelines/publish.yml 仍然是手动 npm 发布流水线
.azure/pipelines/mirror.yml 仍然可用于存储库镜像工作流
仅在本地预发布检查清单和 CI 都在 Node 20 上顺利通过后才发布
在决定更改是否需要 npm、MCP Registry 或仅注册表预发布时，请遵循 RELEASE _POLICY.md

许可证

MIT

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections. Dates show when Glama detected each change.

5 tool updatesv1.0.1Apr 8, 2026
- First observedanalyze_server
- First observedcompare_to_baseline
- First observedget_history
- First observedrecord_baseline
- First observedsnapshot

Install Server

RayrunBuild, deploy, and operate MCP servers without becoming an infrastructure team.ray.runAd

TDQS

B3.4/5.0

Disambiguation4/5

Tools are mostly distinct, though analyze_server and compare_to_baseline both perform explanatory analysis of deviations, which could cause brief hesitation. snapshot vs analyze_server clearly separates raw collection from analysis.

Naming Consistency4/5

Four tools follow a clear verb_noun pattern (analyze_server, get_history, record_baseline, compare_to_baseline). snapshot stands alone without a verb prefix, though it functions as a verb in context—minor deviation but readable.

Tool Count5/5

Five tools is ideal for this focused scope: baseline lifecycle (record, compare), historical retrieval, real-time analysis, and raw capture. Each serves a specific step in the monitoring workflow without bloat.

Completeness4/5

Covers the core monitoring lifecycle: establishing baselines, recording current state, historical lookup, and comparative analysis. Minor gaps exist (no list_baselines or delete_baseline), but agents can work around these.

Resources

Need Help?

Related Servers

Related MCP Connectors

Netdata
io.github.netdata
Real-time infrastructure monitoring with metrics, logs, alerts, and ML-based anomaly detection.
Monitoring Observability
277,495
emisarOAuth
dev.emisar
Let AI operate servers without SSH. Choose actions, approve risky changes, and audit every step.
Cloud Platforms Security Networking & Infrastructure
2354
mcp A
com.googleapis.monitoring
Gain visibility into the performance, availability, and health of your apps and infrastructure.
Monitoring Observability
92
browser-ssh
com.browserssh
Scoped, audited SSH exec, sessions, and SFTP on your saved servers without exposing credentials
Shell Access File Systems
2

Latest Blog Posts

Who's Calling? MCP Hosts Are an Identity Blind Spot (And the Spec Knows It)
By Om-Shree-0709 on July 25, 2026.
mcp
Agent Identity
OAuth 2.1
Your AI Chatbot Just Exposed Your CEO's Salary to an Intern
By Om-Shree-0709 on July 2, 2026.
Agent Identity
MCP Security
OAuth Delegation
Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)
By Om-Shree-0709 on June 30, 2026.
Agentic Ai
Prompt Injection
WebAssembly

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/oaslananka/mcp-infra-lens'

If you have feedback or need assistance with the MCP directory API, please join our Discord server