Conduct fingerprint analysis and bypass solution design for $ARGUMENTS .
Core concept: The anti-fingerprinting system checks thecross-layer consistencyof fingerprints across the four layers: TLS, HTTP/2, TCP/IP, and browser API. Spoofing a single layer is easy, but making all four layers self-consistent is hard—this is the root cause of most solutions failing.
Four-layer fingerprint system
Layer 1: TCP/IP layer
Characteristics exposed by the operating system during the TCP handshake that cannot be spoofed at the application layer:
| Characteristic | Windows | Linux | macOS |
|---|---|---|---|
| TTL | 128 | 64 | 64 |
| TCP Timestamp | None | Yes | Yes |
| Window Size | 65535 | 29200 | 65535 |
| TCP Options order | MSS,NOP,WScale,NOP,NOP,SACK | MSS,SACK,Timestamp,NOP,WScale | MSS,NOP,WScale,NOP,NOP,Timestamp,SACK,EOL |
Key: TCP-layer fingerprints are determined by the operating system kernel, and the application layer cannot control them. UA claims Windows but TTL=64 → immediate contradiction.
Layer 2: TLS layer
Fingerprints in the TLS ClientHello:
| Fingerprint type | Composition | Purpose |
|---|---|---|
| JA3 | cipher_suites + extensions + groups + ec_point_formats | Classic TLS fingerprint |
| JA4 | Protocol version + SNI + cipher count + extension count + ALPN | Next-generation fingerprint |
Chrome characteristics:
- Contains GREASE values (random values such as 0x0a0a, 0x1a1a)
- Extension list includes ECH (encrypted_client_hello)
- Chrome 116+ randomizes extension order (JA3 is no longer stable, JA4 is more reliable)
Node.js characteristics(exposure points):
- No GREASE values
- The cipher suite list differs from browsers
- Missing
application_settingsand other Chrome-specific extensions
Layer 3: HTTP/2 layer
SETTINGS frame parameters when an HTTP/2 connection is established:
| Parameter | Chrome | Firefox | Safari | Node.js (default) |
|---|---|---|---|---|
| HEADER_TABLE_SIZE | 65536 | 65536 | — | 4096 |
| INITIAL_WINDOW_SIZE | 6291456 | 131072 | — | 65535 |
| MAX_HEADER_LIST_SIZE | 262144 | — | — | — |
| WINDOW_UPDATE | 15663105 | 12517377 | 10420225 | Not sent |
| Pseudo-header order | m,a,s,p | m,p,a,s | m,s,a,p | m,a,s,p |
Key: The WINDOW_UPDATE value is the strongest signal for distinguishing browsers, and almost no tool spoofs it correctly.
Layer 4: Browser API layer
Information exposed by JavaScript-accessible APIs:
| Category | Detection item | Common problems with automation tools |
|---|---|---|
| Automation detection | navigator.webdriver |
Selenium/Puppeteer default true |
| Browser identification | window.chrome |
headless Chrome lacks this object |
| Canvas | toDataURL() hash |
headless rendering results differ |
| WebGL | WEBGL_debug_renderer_info |
SwiftShader / Mesa exposes headless |
| Audio | AudioContext output |
Different hardware has different LSBs |
| Permissions | Permissions.query() |
Automation tools return denied |
| CDP detection | CDP traces in error stacks | Puppeteer leaks the CDP call stack |
| Plugins | navigator.plugins |
headless is usually empty |
Analysis process
Step 1: Identify the detection system
常见反指纹服务:
Cloudflare → 检查 cf-ray header、Turnstile 验证
Akamai → 检查 _abck cookie、sensor_data 请求
DataDome → 检查 datadome cookie、interstitial 页面
PerimeterX → 检查 _px cookie
判断方法:
1. 查看 403/429 响应头中的特征
2. 搜索页面 JS 中的特征字符串
3. 分析 cookie 名称和 JS 请求模式
Step 2: Collect the current fingerprint
If fingerprint-collector can be deployed, use it for comprehensive self-testing:
# 部署 fingerprint-collector
git clone https://github.com/zhizhuodemao/fingerprint-collector.git
cd fingerprint-collector && pip install -r requirements.txt
sudo ENABLE_TCP=1 python app.py
# 用你的工具/脚本访问 https://localhost:8443
# 查看 /api/analysis 获取检测报告
When deployment is not possible, public testing sites can be used:
tls.browserleaks.com— TLS Fingerprintbrowserleaks.com/canvas— Canvas Fingerprintbrowserleaks.com/webgl— WebGL Fingerprint
Step 3: Consistency Diagnosis
Check for inconsistencies against the following contradiction detection table:
| Check Item | Consistency Rule | Common Contradictions |
|---|---|---|
| UA ↔ TLS | The browser claimed by the UA matches JA3/JA4 | Node.js TLS fingerprint but UA says Chrome |
| UA ↔ HTTP/2 | SETTINGS parameters are consistent with the UA browser | HTTP/2 uses default values, UA says Chrome |
| UA ↔ TCP | TTL is consistent with the operating system claimed by the UA | Linux server (TTL=64) but UA says Windows |
| WebGL ↔ UA | GPU renderer is consistent with the claimed platform | SwiftShader (headless indicator) |
| Timezone ↔ IP | JS timezone is consistent with IP geolocation | UTC+8 but IP is in the United States |
| Language ↔ IP | navigator.languages is consistent with the IP region | Chinese language but US IP |
Step 4: Develop a Bypass Plan
Choose a plan based on the diagnosis results:
Scenario A: Python requests are blocked
方案 1(推荐): curl_cffi — 模拟真实浏览器 TLS + HTTP/2 指纹
pip install curl_cffi
from curl_cffi import requests
r = requests.get(url, impersonate="chrome")
方案 2: tls_client — Go 实现的 TLS 指纹库
pip install tls_client
session = tls_client.Session(client_identifier="chrome_120")
方案 3: 自定义 TLS — 使用 PyOpenSSL + h2 精确控制
需要手动设置 cipher suites、extensions、HTTP/2 SETTINGS
Scenario B: Automated browser is detected
方案 1: Patchright(推荐) — Playwright 反检测分支
js-reverse-mcp 已内置 Patchright 引擎
方案 2: undetected-chromedriver — Selenium 反检测
pip install undetected-chromedriver
方案 3: Playwright + stealth 插件
npm install playwright-extra puppeteer-extra-plugin-stealth
关键配置:
- 使用持久化 user-data-dir(保持登录态 + cookie)
- 不使用 headless 模式(或使用 new headless)
- 注入 stealth 脚本消除自动化痕迹
Scenario C: Node.js environment emulation is identified
参考 env-patch skill 的 references/fingerprint-consistency.md
重点:
- navigator.webdriver = false(正确的属性描述符)
- window.chrome 对象完整性
- Canvas/WebGL 返回有意义的值
- AudioContext 行为正确
Spoofing Tool Comparison
| Tool | TLS | HTTP/2 | TCP | Browser API | Applicable Scenario |
|---|---|---|---|---|---|
| curl_cffi | ✅ | ✅ | ❌ | ❌ | Pure API requests |
| tls_client | ✅ | ✅ | ❌ | ❌ | Pure API requests |
| Patchright | ✅ | ✅ | ❌ | ✅ | Need to execute JS |
| undetected-chromedriver | ✅ | ✅ | ❌ | ✅ | Selenium scenario |
| Real browser + proxy | ✅ | ✅ | ✅ | ✅ | Highest realism |
TCP layer: Unless running on an operating system consistent with what the UA claims, the TCP layer cannot be spoofed. This is the "ultimate detection" — if the target performs TCP layer detection, you must use a machine with the corresponding OS.
Anti-patterns
- Don't only change the User-Agent — UA is the easiest fingerprint to check and also the least valuable; changing the UA without changing TLS/HTTP2 is equivalent to exposing yourself.
- Don't mix characteristics from different browsers — Chrome's TLS + Firefox's HTTP/2 = a client that does not exist in the world.
- Don't claim Windows on Linux — TCP TTL will expose it.
- Don't ignore HTTP/2 WINDOW_UPDATE — This is the most overlooked but most distinctive fingerprint.
- Don't make requests directly with the requests library — The TLS fingerprint of Python requests is distinctive, and any detection system can identify it.
Completion Criteria
指纹绕过结果:
- 目标站点:$ARGUMENTS
- 检测系统:[Cloudflare | Akamai | DataDome | 其他]
- 矛盾维度:[TLS | HTTP/2 | TCP | 浏览器API | 跨层不一致]
- 绕过方案:[curl_cffi | Patchright | 其他]
- 验证结果:请求返回 200 + 正常业务数据(非验证码/拦截页)
References
| File | When to Read |
|---|---|
references/tls-fingerprint.md |
Need an in-depth understanding of JA3/JA4 fingerprint construction and spoofing details |
references/http2-fingerprint.md |
Need to precisely match HTTP/2 SETTINGS and WINDOW_UPDATE |
references/browser-fingerprint.md |
Need to supplement browser API layer fingerprints (Canvas/WebGL/Audio, etc.) |