AI 에이전트(Claude Code, Claude Desktop, Cursor, Windsurf, Cline, Antigravity 등)를 위한 안티봇 방어 기능을 갖춘 fetch 도구를 제공하는 고성능 Model Context Protocol (MCP) 서버입니다.
@trishchuk/fetch를 기반으로 구동되며, 실제 브라우저의 TLS (JA3/JA4, ClientHello) 및 HTTP/2 핑거프린트를 정확하게 모방하는 네이티브 curl-impersonate 스타일 HTTP 클라이언트입니다.
표준 Node.js/Undici HTTP 클라이언트는 최신 봇 보호 시스템(Cloudflare Turnstile / Under Attack Mode, DataDome, PerimeterX / HUMAN, Akamai, Kasada, AWS WAF)에 의해 즉시 감지되어 차단됩니다.
또한, 표준 MCP 페칭 도구는 대용량 페이로드에서 자주 실패하거나 LLM 토큰 컨텍스트를 폭발시킵니다.
@trishchuk/mcp-fetch-server는 두 가지 문제를 모두 해결합니다:
현실적인 브라우저 가장: 최신 브라우저(Chrome, Safari, Firefox)의 정확한 암호 제품군, TLS 확장, ALPN 순서 및 HTTP/2 설정 프레임을 복제합니다.
LLM 컨텍스트 안전 잘림: 응답 본문을 2MB(maxResponseBytes)로 스트리밍하고 제한합니다. 초과하는 큰 페이지는 오류와 함께 중단되는 대신 깔끔하게 잘리고 "truncated": true로 표시됩니다.
상태 저장 세션: session 매개변수를 사용하여 여러 에이전트 도구 호출 간에 쿠키, 로그인 상태 및 연결 풀을 유지합니다.
스마트 인코딩: MIME 유형을 자동 감지하여 HTML/JSON/XML의 경우 깨끗한 UTF-8 텍스트를, 바이너리 파일(이미지, PDF, 문서)의 경우 Base64를 반환합니다.
Related MCP server: smart-webfetch-mcp
Node.js >= 24 — @trishchuk/fetch에 필요합니다.
사전 빌드된 네이티브 바이너리는 macOS (arm64, x64), Linux (x64/arm64, glibc 및 musl) 및 Windows (x64)용으로 제공됩니다. 다른 플랫폼은 기본 클라이언트에서 지원되지 않습니다.
npx로 실행 (설치 불필요)npx를 통해 서버를 직접 실행할 수 있습니다:
npx -y @trishchuk/mcp-fetch-server# Global
npm install -g @trishchuk/mcp-fetch-server
# Or clone & install locally
git clone https://github.com/x51xxx/mcp-fetch-server.git
cd mcp-fetch-server
npm installCLI를 통해 직접 추가:
# Using npx (recommended)
claude mcp add fetch -- npx -y @trishchuk/mcp-fetch-server
# Or using local path
claude mcp add fetch -- node /path/to/mcp-fetch-server/src/index.jsclaude_desktop_config.json에 추가:
macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json
Linux: ~/.config/Claude/claude_desktop_config.json
{
"mcpServers": {
"fetch": {
"command": "npx",
"args": ["-y", "@trishchuk/mcp-fetch-server"]
}
}
}.mcp.json)작업 공간에 .mcp.json을 생성하거나 업데이트:
{
"mcpServers": {
"fetch": {
"command": "npx",
"args": ["-y", "@trishchuk/mcp-fetch-server"]
}
}
}fetch매개변수 | 유형 | 기본값 | 설명 |
|
| 필수 | 대상 절대 URL (예: |
|
|
| HTTP 메서드 ( |
|
|
| 키-값 쌍의 요청 헤더 ( |
|
|
| UTF-8 문자열로 전송되는 요청 본문 (JSON, 폼 인코딩, 원시 텍스트). |
|
|
| 브라우저 핑거프린트 프리셋 (예: |
|
|
| 선언된 OS: |
|
|
| 프록시 URL: |
|
|
| 여러 호출 간 클라이언트 연결 및 쿠키 저장소 공유를 위한 세션 ID. |
|
|
| 사용자 정의 DNS 핀 (예: |
|
|
| 리디렉션 모드: |
|
|
| 프로토콜 버전 강제: |
|
|
| 최소 TLS 버전: |
|
|
| 최대 TLS 버전: |
|
|
| 밀리초 단위의 전체 요청 시간 제한. |
|
|
| 본문 제한 (최대 2MB). 이를 초과하는 응답은 안전하게 잘립니다. |
|
|
| 본문 반환 형식: |
완료된 모든 HTTP 전송은 표준 JSON 결과를 반환합니다 (redirect: "manual" 아래의 404, 500 또는 3xx 포함):
{
"status": 200,
"statusText": "OK",
"ok": true,
"url": "https://example.com/data",
"redirected": false,
"headers": {
"content-type": "application/json; charset=utf-8",
"cache-control": "max-age=3600"
},
"bodyEncoding": "text",
"body": "{\"message\": \"Hello world\"}",
"truncated": false
}네트워크 연결이 실패하거나 시간이 초과되거나 URL이 유효하지 않은 경우, 도구는 isError: true를 반환합니다:
{
"error": true,
"code": "TIMEOUT",
"message": "failed to read response body: request or response body error: operation timed out"
}{
"url": "https://protected-site.com/products",
"impersonate": "chrome_147",
"platform": "macos",
"headers": {
"Accept-Language": "en-US,en;q=0.9"
}
}// Step 1: Login / Obtain Session Cookie
{
"url": "https://example.com/api/login",
"method": "POST",
"session": "agent-crawler-01",
"headers": { "Content-Type": "application/json" },
"body": "{\"user\":\"admin\",\"password\":\"secret\"}"
}
// Step 2: Access protected resource (session cookies automatically preserved)
{
"url": "https://example.com/api/dashboard",
"session": "agent-crawler-01"
}{
"url": "https://geo-restricted.example.com",
"proxy": "socks5://user:pass@proxy.example.com:1080",
"impersonate": "safari_26"
}{
"url": "https://example.com/report.pdf",
"encoding": "base64"
}{
"url": "https://internal-origin.example.com/feed",
"resolve": {
"internal-origin.example.com": "192.0.2.42"
},
"redirect": "manual"
}@trishchuk/mcp-fetch-server는 광범위한 브라우저 핑거프린트를 지원합니다:
Chrome: "chrome_100" … "chrome_149" (예: "chrome_147", "chrome_131", "chrome_116")
Edge: "edge_101" … "edge_148"
Opera: "opera_116" … "opera_131"
Firefox: "firefox_109", "firefox_133", "firefox_147" …, "firefox_private_136" 및 "firefox_android_135" 포함
Safari: "safari_15.3" … "safari_26.4", iOS/iPad 변형 포함 ("safari_ios_26", "safari_ipad_26")
OkHttp (Android 앱): "okhttp_3.9" … "okhttp_5"
동적: "random", "weighted_random" (핑거프린트를 자동 순환, session별 고정)
버전 번호는 밑줄을 사용합니다 (chrome_147, chrome147 아님). 알 수 없는 이름은 모든 허용된 변형을 나열하는 InvalidArg 오류와 함께 빠르게 실패합니다.
npm install
npm start # run the server over stdio
npm test # end-to-end smoke tests, no network required
npm run format # format with Biome
npm run lint # lint with Biome
npm run check # format + lint check, also run before publish테스트는 stdio를 통해 실제 서버를 실행하고 로컬 HTTP 서버에 대해 MCP 클라이언트로 구동하여 제한에서의 잘림, 리디렉션 모드, HEAD, base64 본문, 시간 초과 및 전송 오류를 다룹니다.
MIT © Taras Trishchuk
HTTP fetch backed by @trishchuk/fetch: a curl-impersonate-style client that emulates a real browser TLS/HTTP2 fingerprint (JA3/JA4, ClientHello, ALPN) so requests are not flagged by fingerprint-based bot detection (Cloudflare, DataDome, PerimeterX, etc.) the way Node's default HTTP client is. Use it for GET/POST/etc. against sites that block or challenge plain scrapers.
| Name | Required | Description | Default |
|---|---|---|---|
| url | Yes | Absolute URL to request. | |
| body | No | Request body, sent as UTF-8 text (e.g. JSON string, form-encoded string). | |
| proxy | No | Proxy URL: http://, https://, or socks5://, optionally with user:pass@. | |
| method | No | HTTP method. | GET |
| headers | No | Request headers. | |
| resolve | No | Hostname-to-IP pinning, e.g. { "example.com": "192.0.2.1" }. Useful to pin DNS for SSRF-safety or A/B hosts. | |
| session | No | Opaque session id. Reusing it across calls keeps the same underlying client and cookie jar (e.g. to stay logged in). | |
| encoding | No | How to return the body: "auto" picks text for text-like content-types and base64 otherwise. | auto |
| platform | No | Declared OS for the fingerprint. | |
| redirect | No | Redirect handling. Defaults to "follow". | |
| timeoutMs | No | Request deadline in milliseconds. | |
| httpVersion | No | Force HTTP/1.1 or HTTP/2 instead of negotiating. | |
| impersonate | No | Browser fingerprint profile, e.g. "chrome_147", "safari_26", "random". Defaults to the library default. | |
| tlsMaxVersion | No | ||
| tlsMinVersion | No | ||
| maxResponseBytes | No | Response body cap in bytes (max 2097152, i.e. 2MB, to keep tool output usable). A larger body is truncated to this size and flagged with "truncated": true, not rejected. |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
The description discloses the critical behavioral trait of browser fingerprint emulation (JA3/JA4, TLS, HTTP/2) to avoid detection. Without annotations, this provides transparency about why requests succeed. However, it does not mention whether the tool is read-only or has side effects, though HTTP fetch is inherently non-destructive to local state.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is extremely concise: two sentences that front-load the core purpose and differentiator, followed by usage guidance. Every sentence serves a purpose, and there is no redundancy or filler.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given the tool has 16 parameters, no annotations, and no output schema, the description is somewhat incomplete. It does not explain what the tool returns (e.g., status code, headers, body format) or that the response is a standard HTTP response. While the schema covers some constraints like maxResponseBytes, the agent lacks clarity on what to expect after invocation.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 88%, so the input schema already documents most parameters. The description adds general context about the underlying library and fingerprinting motivation but does not provide additional details on individual parameters beyond what schema descriptions offer. The baseline of 3 is appropriate.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly identifies the tool as an HTTP fetch client with browser fingerprint emulation to bypass bot detection. It specifies the verb 'fetch', the resource (URLs), and the unique value proposition (curl-impersonate style). This distinguishes it from standard HTTP clients and makes its purpose immediately obvious.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description explicitly states when to use the tool: 'against sites that block or challenge plain scrapers'. It contrasts with Node's default HTTP client, implying an alternative. However, it does not list explicit sibling tools or provide when-not-to-use guidance, such as for sites without bot detection.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
v0.1.0fetchScored across 1 tool
With only a single tool, there is no possibility of confusion with other tools. The tool's purpose is clearly described and distinct by default.
With only one tool, there is no pattern to judge. The name 'fetch' is simple and conventional, matching common HTTP client terminology, but the lack of a verb_noun convention is neutral.
A single tool for an HTTP fetch server is too minimal. Most similar servers would include additional tools for managing headers, cookies, or caching, making this feel under-scoped for its stated purpose of scraping.
The server only provides a basic fetch tool with no support for managing sessions, handling redirects, managing cookies, or performing other common HTTP operations. This leaves significant gaps for any realistic scraping workflow.
Reliable web fetching for AI agents with retry, circuit breaker, caching, and anti-bot bypass
Reliable web access for AI agents: smart HTTP, rotating proxies, and full-browser rendering.
Prompt-injection scanning and safe webpage fetching for AI agents reading untrusted content.
Web search, browser automation, scraping, crawling and CAPTCHA solving for AI agents.
