web-mcp
Provides academic paper search using the arXiv search engine.
Provides web search using Baidu, especially effective for Chinese-language queries.
Provides web search using Brave Search with rate limiting and retry handling.
Provides technical content search using CSDN, a Chinese developer community.
Provides web search using DuckDuckGo with anti-bot protections.
Provides code and repository search, plus automatic fetching of GitHub README files.
Provides Google search via the Serper API, offering stable and reliable results (requires SERPER_API_KEY).
Provides web search via a self-hosted SearXNG instance, supporting multiple meta-search backends.
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@web-mcpsearch and summarize the latest AI research papers"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
web-mcp
联网搜索 + 网页抓取的 MCP 服务。多引擎并行搜索(免 key)、正文提取、搜索摘要与深度搜索,支持 stdio 与 Streamable HTTP 两种部署方式。任何支持 MCP 的 agent(Claude Code、Cursor、Codex、pi 等)都能接入。
设计上吸收了多个主流开源 MCP 项目的优点,并针对免费搜索引擎的反爬做了大量实测验证。
借鉴项目 | 吸收的特性 |
多引擎并行编排、失败容忍、反爬关键词检测、CSDN 垂直引擎、GitHub README 抓取、代理支持 | |
多格式抓取(text / readable / json)、自定义请求头 | |
搜索引擎限流保护、结构化结果输出 | |
可选 SearXNG 自托管引擎 | |
可选 Tavily API 引擎 |
特性
免 key 优先:baidu / brave / bing / duckduckgo / csdn / github / arxiv 免 key 可用;可选 serper(Google)、tavily、searxng
多引擎并行:
Promise.allSettled编排,单引擎失败不阻塞整体,结果按 URL 去重并带engine来源标签反爬全套应对(实测验证):验证页检测、浏览器渲染兜底、brave 429 节流退避、ddg 202 转 POST、cookie jar、代理自动生效
意图感知路由:按查询特征(code / academic / news / general)自动选择引擎组合
站点权重重排:官方文档加权、内容农场降权,黑名单剔除,支持环境变量覆盖
网页抓取三模式:text / readable(Readability)/ json,另支持 PDF、DOCX 解析与 GitHub README 自动抓取
搜索增强:
search_summarize(搜索→抓正文→LLM 带引用摘要)与deep_search(多角度子查询)两种部署:stdio(本地 MCP 客户端)与 Streamable HTTP(联网部署,可选 Bearer 鉴权)
Related MCP server: webmcp
四个工具
1. web_search — 多引擎并行搜索
免 key 引擎并行执行,每个引擎满额请求,结果按 URL 去重合并,每条带 engine 来源标签。
时间盒竞速:默认 5 秒(WEB_MCP_TIME_BOX 可调)。结果数达标(≥ max_results)立即取消其余引擎提前返回;到期未达标也返回已有结果,绝不等待最慢引擎。返回 elapsedMs(实际耗时)、waived(因时间盒放弃的引擎)、timeBoxMs。
意图感知路由(auto 模式):
意图 | 中文查询 | 英文查询 |
code(api/报错/教程/框架) | csdn + github + brave + baidu | github + brave + bing |
academic(论文/研究) | baidu + arxiv + brave + csdn | arxiv + brave + bing |
news(发布/新闻/最新) | baidu + brave + bing | brave + bing + ddg |
general | baidu + csdn + brave + ddg | brave + baidu + bing + ddg |
站点权重重排:内置权重表(github.com ×1.6、官方文档站加权;百家号 ×0.55、内容农场降权),黑名单域名直接剔除。可用 SITE_WEIGHTS(JSON)/ SITE_BLOCKLIST(逗号分隔)覆盖。
recency 时间过滤:recency=day/week/month/year 透传各引擎(brave tf、bing freshness、ddg df、serper tbs)。
2. fetch_page — 网页抓取(三模式 + 文档解析 + 浏览器兜底)
模式 | 说明 |
| 粗提取为 markdown,保留链接格式,截断到 |
| Mozilla Readability 提取正文,去导航/广告(适合文章页),失败自动降级 text |
| 直接解析 JSON API 接口( |
附加能力:
PDF / DOCX 自动解析(pdf-parse + mammoth,30MB 上限)
GitHub 仓库页自动抓 README(raw.githubusercontent.com,走代理)
403/5xx 时浏览器渲染兜底(知乎等反爬站实测可抓)
自定义请求头(
headers参数,如带 Cookie 绕过登录墙)
3. search_summarize — 搜索 + 摘要
一步完成:搜索 → 并行抓 top 结果正文 → DeepSeek 生成带 [1][2] 引用标注的结构化中文摘要(总述 + 要点 + 关键事实)。查资料写报告场景省去多次 search+fetch 往返。无 DEEPSEEK_API_KEY 时退化为返回搜索+抓取结果。
4. deep_search — 多角度深度搜索
LLM 把查询拆成多个角度(官方/技术细节/评测/最新动态…)的子查询并行搜索,分组返回。无 LLM 时用规则拆分(原查询 + 官网 + 教程)。
快速开始
git clone https://github.com/Chang-Tong/web-mcp.git
cd web-mcp
npm install
# stdio 模式(本地 MCP 客户端)
node src/index.mjs
# HTTP 模式(联网部署,局域网/公网可访问)
node src/index.mjs --http 8787
# 带鉴权启动(推荐公网部署时开启)
WEB_MCP_AUTH_TOKEN=your-secret node src/index.mjs --http 8787接入客户端
Claude Code / Codex(~/.claude/settings.json 或项目 .mcp.json):
{
"mcpServers": {
"web": {
"command": "node",
"args": ["/absolute/path/to/web-mcp/src/index.mjs"]
}
}
}Cursor(Settings → MCP,远程模式):
{
"mcpServers": {
"web": {
"url": "http://localhost:8787/mcp",
"headers": {
"Accept": "application/json, text/event-stream",
"Authorization": "Bearer your-secret"
}
}
}
}curl 验证 HTTP 服务:
curl http://localhost:8787/health # 健康检查,返回 { ok: true, ... }配置(环境变量)
变量 | 默认 | 说明 |
|
|
|
| — | Serper 的 Google 搜索 key(免费 2500 次/月),最稳定,推荐配置 |
|
| Serper 搜索地区/语言 |
| — | Tavily key(可选引擎) |
| — | 自托管 SearXNG 实例地址(可选引擎,需开启 JSON 格式) |
| 自动读取 | search_summarize / deep_search 的 LLM(优先环境变量,其次 pi 的 |
|
| 可指向 OpenAI 兼容的任意端点 |
|
| LLM 模型名 |
| 内置表 | JSON 格式域名→权重,如 |
| — | 逗号分隔的域名黑名单(直接剔除),如 |
| 自动 |
|
| 自动探测 | 指定浏览器可执行文件路径(系统 Chrome/Edge、playwright 缓存自动探测) |
| — | HTTP 模式鉴权,设置后要求 |
| — | 设为 |
|
| web_search 时间盒(ms):达标提前收 / 到期返回已有结果,不等最慢引擎 |
代理自动生效:http_proxy / https_proxy / no_proxy(undici EnvHttpProxyAgent),无代理环境自动直连,无需额外配置。
架构
MCP 客户端 (Claude Code / Cursor / Codex / ...)
│ stdio 或 Streamable HTTP(stateful session,CORS + 可选 Bearer 鉴权)
▼
web-mcp (Node + @modelcontextprotocol/sdk)
├── web_search ── 并行多引擎 ──┬─ baidu / csdn(中文,免 key)
│ ├─ brave(免 key,节流+退避重试)
│ ├─ bing / duckduckgo(免 key)
│ └─ serper / tavily / searxng(可选)
└── fetch_page ── text / readable(Readability) / json / GitHub README已知限制
免费引擎在代理 IP 池下会被间歇风控(brave 429、baidu 验证页、ddg 202),多引擎并行 + 快速失败 + 浏览器兜底已把影响降到最低;要最稳定体验请配置
SERPER_API_KEY浏览器兜底需要本机有 Chrome/Edge/Chromium(自动探测系统浏览器与 playwright 缓存),首次启动浏览器约 3-10s
search_summarize / deep_search 依赖
DEEPSEEK_API_KEY(默认自动读 pi 凭据)
测试
node test-client.mjs "搜索词" # stdio 冒烟测试:搜索 + 抓取贡献
欢迎提交 Issue 和 PR。代码风格保持简单,每个引擎的失败容忍逻辑建议附上实测场景说明。
许可证
This server cannot be installed
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Related MCP Servers
- AlicenseAqualityAmaintenanceMCP server for private web search via self-hosted SearXNG with local reranking, full-page content fetching via Firecrawl, and optional Ollama-powered query expansion and summaries.Last updated729316MIT
- Alicense-qualityCmaintenanceMCP server for web search and content extraction using DuckDuckGo or SearXNG, with Playwright-based fetching and LLM-powered data extraction.Last updated140MIT
- Alicense-qualityDmaintenanceMCP server for internet search via direct Google and DuckDuckGo HTML scraping with AI-powered result normalization and optional summarization, requiring no API keys for search.Last updatedMIT
- Alicense-qualityBmaintenanceA zero-config web search and fetch MCP server for LLM agents, featuring multi-backend metasearch, persistent rolling cache, and structured error envelopes for retry-friendly interactions.Last updatedMIT
Related MCP Connectors
MCP server for Google search results via SERP API
Multi-engine search for AI agents. Trust scoring, local corpus, MCP-native. Self-hostable, BYOK.
Fast, intelligent web search and web crawling. New mcp tool: Exa-code is a context tool for coding
Latest Blog Posts
- Who's Calling? MCP Hosts Are an Identity Blind Spot (And the Spec Knows It)By Om-Shree-0709 on .mcpAgent IdentityOAuth 2.1
- Your AI Chatbot Just Exposed Your CEO's Salary to an InternBy Om-Shree-0709 on .Agent IdentityMCP SecurityOAuth Delegation
- Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)By Om-Shree-0709 on .Agentic AiPrompt InjectionWebAssembly
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/Chang-Tong/web-mcp'
If you have feedback or need assistance with the MCP directory API, please join our Discord server