mcpField
H
httpx-probe

httpx-probe

github.com/wgpsecSecurity1.8k50 installsUpdated 4d ago
Repository

About httpx-probe

httpx-probe is a security skill guide for using ProjectDiscovery’s httpx to check HTTP availability and fingerprint web targets. It covers batch probing, response filtering, screenshots, and pipelines with other ProjectDiscovery tools.

Key features

  1. 01HTTP availability checks
  2. 02Status, title, and technology detection
  3. 03Response matching and filtering
  4. 04Headless page screenshots
  5. 05Pipelines with scanning tools

Use cases

  1. 01Checking many URLs for live HTTP services
  2. 02Identifying web technologies across targets
  3. 03Capturing screenshots of target pages
  4. 04Passing discovered hosts into vulnerability scans

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/wgpsec/aboutsecurity --skill httpx-probe

After installing, type/httpx-probein Claude Code to confirm it loaded.

Bundled filesevals/evals.json

Details

From the project README

httpx HTTP 探活与指纹识别方法论

httpx 是 ProjectDiscovery 出品的多功能 HTTP 探针工具。核心优势:高速并发(默认 50 线程)+ 丰富探针(状态码/标题/技术栈/CDN/截图等)+ 管道核心(连接端口扫描和漏洞扫描的桥梁)。

项目地址:https://github.com/projectdiscovery/httpx

Phase 1: 基本存活检测

# 单个目标
httpx -u http://target.com

# 批量目标(从文件)
httpx -l urls.txt

# 从 stdin 管道
cat urls.txt | httpx

# 静默输出(只显示存活 URL)
cat urls.txt | httpx -silent

Phase 2: 信息探针(核心能力)

# 显示状态码 + 标题 + 技术栈(最常用组合)
httpx -l urls.txt -sc -title -tech-detect

# 显示状态码 + 内容长度 + Web 服务器
httpx -l urls.txt -sc -cl -server

# 显示 IP + CNAME + ASN
httpx -l urls.txt -ip -cname -asn

# CDN/WAF 检测
httpx -l urls.txt -cdn

# 响应时间
httpx -l urls.txt -rt

# 完整信息输出
httpx -l urls.txt -sc -title -tech-detect -server -ip -cdn -cl

Phase 3: 技术栈识别(指纹识别)

httpx 内置 Wappalyzer 数据集,能识别 Web 框架、CMS、编程语言等:

# 技术栈检测
httpx -u http://target.com -tech-detect

# 输出示例:
# http://target.com [200] [WordPress] [PHP,MySQL,Apache]

# 批量识别 + JSON 输出
httpx -l urls.txt -tech-detect -json -o fingerprints.json

# 自定义指纹文件
httpx -l urls.txt -tech-detect -custom-fingerprint-file /path/to/fingerprints.json

Phase 4: 过滤与匹配

# 只显示 200 OK 的目标
httpx -l urls.txt -mc 200

# 排除 404 和 403
httpx -l urls.txt -fc 404,403

# 匹配包含特定字符串的响应
httpx -l urls.txt -ms "admin"

# 正则匹配
httpx -l urls.txt -mr "login|admin|dashboard"

# 按响应长度过滤
httpx -l urls.txt -fl 0

# 匹配特定 CDN
httpx -l urls.txt -match-cdn cloudflare

# 过滤错误页面
httpx -l urls.txt -filter-page-type error,parked

Phase 5: 截图(Headless)

# 页面截图
httpx -l urls.txt -screenshot

# 使用系统 Chrome
httpx -l urls.txt -screenshot -system-chrome

# 截图 + 技术栈 + 状态码
httpx -l urls.txt -screenshot -tech-detect -sc -title

Phase 6: 管道集成

httpx 是 ProjectDiscovery 工具链的核心枢纽:

# 子域名 → HTTP 存活
subfinder -d target.com -silent | httpx -silent

# 端口扫描 → HTTP 存活 → 指纹
naabu -host target.com -silent | httpx -tech-detect -title -sc

# 子域名 → 存活 → 漏洞扫描
subfinder -d target.com -silent | httpx -silent | nuclei -severity critical,high

# 子域名 → 存活 → 爬虫 → 漏洞扫描(完整链)
subfinder -d target.com -silent | httpx -silent | katana -jc -silent | nuclei

# 网段扫描 → HTTP 存活 → 指纹 → 高价值目标筛选
naabu -host 10.0.0.0/24 -p 80,443,8080 -silent | httpx -tech-detect -title -sc -silent

Phase 7: 高级用法

# 探测多端口(单个主机的多端口 HTTP 检测)
echo target.com | httpx -ports 80,443,8080,8443

# 自定义请求方法
httpx -l urls.txt -x GET,POST

# 自定义 Header
httpx -l urls.txt -H "Cookie: session=abc"

# 跟随重定向
httpx -l urls.txt -follow-redirects

# TLS 信息抓取
httpx -l urls.txt -tls-grab

# 提取正则匹配内容(如提取邮箱)
httpx -l urls.txt -extract-preset mail,ipv4

# 自定义正则提取
httpx -l urls.txt -extract-regex "api[_-]?key['\"]?\s*[:=]\s*['\"]?([a-zA-Z0-9]+)"

# JSON 输出(包含完整信息)
httpx -l urls.txt -json -o results.json

# 控制并发和速率
httpx -l urls.txt -threads 100 -rate-limit 50

# 去重(过滤相似页面)
httpx -l urls.txt -filter-duplicates

常用场景速查

场景命令
快速存活检测cat urls.txt | httpx -silent
技术栈识别httpx -l urls.txt -tech-detect -sc -title
找管理后台httpx -l urls.txt -mr "admin|login|dashboard" -sc -title
CDN 识别httpx -l urls.txt -cdn -ip
批量截图httpx -l urls.txt -screenshot -system-chrome
提取子域名httpx -l urls.txt -extract-fqdn -json
连接 naabunaabu -host target -silent | httpx -tech-detect

Comments

Markdown supported