Network MCP (Labeveryday)

网络MCP服务器

模型上下文协议（MCP）服务器，为AI代理提供网络诊断工具。旨在将繁重的网络分析卸载到服务器上，并返回针对LLM消费优化的结构化、可操作的数据。

](https://pypi.org/project/network-mcp/)  

特性

• 连接性测试：ping、跟踪路由、DNS查找、端口检查、MTR
• 批量操作：同时测试多个主机/端口
• 本地网络信息：获取接口、路由、DNS配置、ARP表、连接（跨平台）
• Pcap分析：使用scapy分析数据包捕获（不需要tshark）
• 自定义过滤器：执行高级查询的scapy过滤器表达式（AST验证）
• 安全控制：用于目标验证的可配置列表/块列表（在连接工具上强制执行）
• PCAP路径护栏：限制允许服务器从哪些目录读取捕获
• 智能摘要：返回人类可读的摘要和结构化数据

安装

pip install network-mcp

或者从源代码安装：

git clone https://github.com/labeveryday/network-mcp.git
cd network-mcp
pip install -e .

快速开始

运行MCP服务器：

network-mcp

或者用Python：

python -m network_mcp.server

建议首次运行（帮助代理决定此主机上的安全/可用内容）：

• 检查功能：呼叫 capabilities 查看已安装的二进制文件（如 mtr)、主动安全策略和PCAP路径护栏。
• 决定政策：如果您需要放松/收紧目标验证或PCAP访问，请在启动前设置env变量。
• 默认情况下运行单元测试: pytest 除非您选择加入，否则跳过集成测试（请参阅开发）。

可用工具

诊断

规划（纯CIDR/VLAN数学）

专为一级/二级NOC工作流程而设计。这些工具是 纯净 （无网络调用，确定性输出），可安全用于规划和验证。

输入示例

• VLAN映射（接受两种格式）：

{
  "10": "192.168.10.0/24",
  "20": { "cidr": "192.168.20.0/24", "name": "Voice" }
}

• plan_subnets 要求（支持别名： hosts 和 prefix):

[
  { "vlan_id": 10, "name": "Users", "hosts": 120 },
  { "vlan_id": 20, "name": "Voice", "hosts": 60 },
  { "vlan_id": 30, "name": "Printers", "prefix": 26 }
]

外部Intel

连接工具

批量操作

本地网络信息工具

适用于Linux、macOS和Windows的跨平台工具。

Pcap分析工具

IDE集成

克劳德桌面版

添加 ~/Library/Application Support/Claude/claude_desktop_config.json （macOS）或 %APPDATA%\Claude\claude_desktop_config.json （Windows）：

{
  "mcpServers": {
    "network-tools": {
      "command": ["network-mcp"]
    }
  }
}

光标

添加到MCP设置中：

{
  "mcpServers": {
    "network-tools": {
      "command": ["network-mcp"]
    }
  }
}

使用紫外线

如果您安装了uv：

{
  "mcpServers": {
    "network-tools": {
      "command": "uv",
      "args": ["run", "--directory", "/path/to/network-mcp", "network-mcp"]
    }
  }
}

示例响应

拼

{
  "success": true,
  "target": "google.com",
  "resolved_ip": "142.250.80.46",
  "packets_sent": 4,
  "packets_received": 4,
  "packet_loss_percent": 0.0,
  "min_latency_ms": 11.2,
  "avg_latency_ms": 12.8,
  "max_latency_ms": 15.1,
  "summary": "google.com is reachable. 4/4 packets received, avg latency 12.8ms"
}

批量Ping

{
  "success": true,
  "total_targets": 3,
  "successful": 3,
  "failed": 0,
  "results": [
    {"target": "8.8.8.8", "success": true, "avg_latency_ms": 12.5},
    {"target": "1.1.1.1", "success": true, "avg_latency_ms": 8.2},
    {"target": "google.com", "success": true, "avg_latency_ms": 15.1}
  ],
  "summary": "Batch ping: 3/3 targets reachable"
}

TCP问题检测

{
  "success": true,
  "file_path": "/tmp/capture.pcap",
  "total_tcp_packets": 15234,
  "issues": [
    {
      "issue_type": "retransmission",
      "count": 47,
      "severity": "medium",
      "recommendation": "Retransmissions indicate packet loss. Check for network congestion."
    }
  ],
  "has_issues": true,
  "summary": "TCP issues detected in 15234 packets: 47 retransmissions"
}

吞吐量（来自PCAP）

{
  "success": true,
  "file_path": "/tmp/capture.pcap",
  "total_packets_scanned": 100000,
  "conversations_analyzed": 87,
  "top_n": 3,
  "sort_by": "mbps_total",
  "conversations": [
    {
      "src_ip": "10.0.0.10",
      "src_port": 51544,
      "dst_ip": "93.184.216.34",
      "dst_port": 443,
      "protocol": "TCP",
      "packets_total": 1240,
      "bytes_total": 18423333,
      "duration_seconds": 9.84,
      "start_time": 1734567890.01,
      "end_time": 1734567899.85,
      "packets_forward": 820,
      "bytes_forward": 17600000,
      "packets_reverse": 420,
      "bytes_reverse": 823333,
      "mbps_forward": 14.308,
      "mbps_reverse": 0.669,
      "mbps_total": 14.977,
      "direction": "10.0.0.10:51544 -> 93.184.216.34:443"
    }
  ],
  "summary": "Throughput analysis: 87 conversations from 100000 packets. Top flow 10.0.0.10:51544 -> 93.184.216.34:443 at ~14.977 Mbps over 9.84s"
}

获取接口

{
  "success": true,
  "interfaces": [
    {
      "name": "en0",
      "status": "up",
      "mac_address": "00:11:22:33:44:55",
      "ipv4_addresses": ["192.168.1.100"],
      "ipv6_addresses": ["fe80::1"],
      "netmask": "255.255.255.0",
      "mtu": 1500
    }
  ],
  "default_interface": "en0",
  "summary": "Found 5 interfaces (3 up). Primary: en0"
}

获取公共IP

{
  "success": true,
  "public_ip": "203.0.113.42",
  "service_used": "ipify.org",
  "summary": "Public IP: 203.0.113.42 (via ipify.org)"
}

RDAP查找（WHOIS风格）

{
  "success": true,
  "query": "1.1.1.1",
  "query_type": "ip",
  "rdap_url": "https://rdap.org/ip/1.1.1.1",
  "handle": "NET-1-1-1-0-1",
  "country": "AU",
  "start_address": "1.1.1.0",
  "end_address": "1.1.1.255",
  "summary": "RDAP 1.1.1.1: 1.1.1.0–1.1.1.255 (AU), handle NET-1-1-1-0-1"
}

ASN查找

{
  "success": true,
  "ip": "1.1.1.1",
  "asn": "13335",
  "prefix": "1.1.1.0/24",
  "country": "AU",
  "registry": "apnic",
  "allocated": "2011-08-11",
  "as_name": "CLOUDFLARENET",
  "summary": "1.1.1.1 originates from AS13335 (CLOUDFLARENET), prefix 1.1.1.0/24"
}

配置

创建 config.yaml 在您的工作目录中或 ~/.network-mcp/config.yaml:

security:
  # Only allow these targets (glob patterns, CIDR ranges)
  allowed_targets:
    - "*.company.com"
    - "10.0.0.0/8"
    - "192.168.0.0/16"

  # Block these targets
  blocked_targets:
    - "*.gov"
    - "localhost"
    - "127.0.0.0/8"

  # Block private IPs
  block_private: false

  # Block cloud metadata endpoints (AWS, GCP, etc.)
  block_cloud_metadata: true

pcap:
  max_packets: 100000
  allow_custom_filters: true
  # Restrict which directories the server is allowed to read pcaps from.
  # (Paths are resolved before checking.)
  allowed_paths:
    - "."
    - "~/Documents"
    - "/tmp"

环境变量：

NETWORK_MCP_ALLOWED_TARGETS="*.company.com,10.0.0.0/8"
NETWORK_MCP_BLOCKED_TARGETS="*.gov,localhost"
NETWORK_MCP_BLOCK_PRIVATE="true"
NETWORK_MCP_MAX_PACKETS="50000"
NETWORK_MCP_PCAP_ALLOWED_PATHS=".,~/Documents,/tmp"

为什么选择MCP for Network Tools？

代币效率：LLM有上下文限制。服务器进行繁重的处理（解析10万个数据包），并返回简洁的摘要而不是原始数据。

更好的推理：法学硕士擅长决策 *什么* 进行调查，而不是解析原始输出。结构化数据有助于做出更好的决策。

一致性：服务器端处理是确定性的。您不需要每次都依赖LLM来正确解释traceroute输出。

例子

这 examples/ 目录包含使用以下命令的工作示例 Strands代理商:

快速启动：

cd examples
pip install strands-agents strands-agents-tools 'strands-agents[ollama]'
python ollama_agent.py

看 examples/README.md 获取完整文档。

发展

# Clone and install dev dependencies
git clone https://github.com/labeveryday/network-mcp.git
cd network-mcp
pip install -e ".[dev]"

# Run unit tests (integration tests are skipped by default)
pytest

# Run integration tests (requires system tools and/or network access)
pytest -m integration

# Run linting
ruff check .

项目结构

network-mcp/
├── src/network_mcp/
│   ├── __init__.py
│   ├── server.py           # FastMCP server
│   ├── config.py           # Configuration and security
│   ├── tools/
│   │   ├── connectivity.py # ping, traceroute, dns, port_check, mtr, batch ops
│   │   ├── local.py        # local network info (interfaces, routes, etc.)
│   │   └── pcap.py         # pcap analysis tools
│   └── models/
│       └── responses.py    # Pydantic response models
├── tests/
├── pyproject.toml
└── README.md

需求

• Python 3.10+
• 系统工具： ping, traceroute （大多数系统的标准配置）
• 可选： mtr MTR工具

许可证

MIT许可证-请参阅 许可证 文件以获取详细信息。