Atlas / MCP servers / awslew / web-search-mcp

web-search-mcpSAFE

Name: web-search-mcp
Author: awslew

mcp/awslew/web-search-mcp

MCP server by awslew · llm · mcp · mcp server · Audited 2026-10-11

⚑ Report / re-scan Source ↗

编码 agent 查资料缺搜索 API key?免 key 搜多个引擎、优先可靠来源并抽取可读正文。 Search the web and read pages without an API key.

Verdict

SAFE

Grade

Trust score

89 /100

Exposed tools

3 3r · 0w · 0d

Transport

stdio

License

MIT

Stars

Overview

From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.

编码 agent 要查最新文档或排查报错,却没有搜索 API key;普通搜索结果还常被广告和低质页面占满。这个工具让 agent 不用 key 也能联网查资料:同时搜索多个引擎,尽量把官方文档等可靠结果排前面,再提取可读的网页正文,省去你手工搜索和复制粘贴。

它以 MCP(Model Context Protocol)stdio 服务接入编码客户端;下文再说明多引擎融合、排序和正文抽取的技术细节。

[](#环境要求) [](#许可证) [](#客户端接入)

为什么选择本项目

通用搜索引擎直接喂给大模型常存在三大痛点:商业推广与 UGC 营销页挤占前排、重定向与泛首页因词面复述取得虚高得分、网页抽取丢失代码块与表格结构。

本项目通过多层质量管线解决上述问题:

多引擎直连与融合:中文查询并行聚合国内直连引擎(百度移动端、必应中国、搜狗),英文查询走国际引擎并行融合,经 RRF(Reciprocal Rank Fusion)多路交叉校验。
两级垃圾域过滤:对垃圾域名硬过滤(remove)、对百家号/知乎/CSDN 等 UGC 域降权(lower)。代码内置一份兜底名单(垃圾 TLD + 短链域 + 策划降权域),开箱即生效;完整名单(4,300+ 规则)由 npm run update:blacklist 从公开上游列表本地生成 —— 该数据文件因上游为 GPL-3.0 许可而不随本仓库分发。实测真实场景查询 nDCG@5 提升 8.4% 且技术查询零回归。
权威源信号加权与官方文档召回:识别官方文档子站(如 docs.*、dev.*)并给予权威加权。离线评估集(40 条标注查询)总体 nDCG@5 由 0.7788 提升至 0.9313,hit@1 由 0.6667 提升至 0.9000;另有 8 条查询的同一份候选池在线开/关对照:信号开启时 hit@1 8/8、关闭时 2/8。⚠️ 该 8 条是按"官方域在候选池内"挑出的,结果偏乐观,离线涨幅也相对 09-12 的旧缓存基线,不宜当作普适增益。对技术查询自动推断官方域并发起站内召回,确保官方文档页进候选池。
API 抢救闸门:仅在生产引擎受限(候选池 raw < 12)时自动并入搜索 API(支持 Tavily / 博查 / 智谱),无 key 时保持纯免费直连。受控 A/B 实测(40 次 API 调用):中文 nDCG@5 0.6359 → 0.7370(+0.1011),英文逐位零影响;对照的"无条件全并入"策略虽然中文同分,却把英文打掉 5.1% —— 这正是设置闸门条件的意义。
结构化 Markdown 抽取:采用 linkedom + Readability + turndown 抽取正文,保留标题、代码块、GFM 表格与链接,内置 GBK/GB2312 编码自动探测与解码。

环境要求

Node.js ≥ 22.5(推荐 Node 24 LTS)
原因:项目内置的持久化查询缓存使用了 Node 原生 node:sqlite 的 DatabaseSync 同步接口,无需额外编译 C++ 原生模块。

快速安装

方式 A:一行式接入(推荐,无需 clone、无需配置路径)

MCP 客户端配置里直接用 Release 包地址,npx 会自动拉取并运行:

{
"mcpServers": {
"web-search": {
"command": "npx",
"args": [
"-y",
"https://github.com/awslew/web-search-mcp/releases/download/v1.2.0/web-search-mcp-1.2.0.tgz"
]
}
}
}

首次运行会拉取依赖(实测约 *30 秒 / 33 MB

Read from source at commit 408160df48a7OBSERVED · 2026-10-11

Connect

Built from this server's own package name, version and transport as found in its source — not copied from anyone's documentation, so it cannot drift against a page we do not control. Replace the environment placeholders with a token scoped to the least it needs.

claude-code

claude mcp add web-search-mcp --env BOCHA_API_KEY=${BOCHA_API_KEY} --env OFFICIAL_DOCS_PASS=${OFFICIAL_DOCS_PASS} --env TAVILY_API_KEY=${TAVILY_API_KEY} --env ZHIPU_API_KEY=${ZHIPU_API_KEY} -- npx -y [email protected]

claude-desktop

{
  "mcpServers": {
    "web-search-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "[email protected]"
      ],
      "env": {
        "BOCHA_API_KEY": "${BOCHA_API_KEY}",
        "OFFICIAL_DOCS_PASS": "${OFFICIAL_DOCS_PASS}",
        "TAVILY_API_KEY": "${TAVILY_API_KEY}",
        "ZHIPU_API_KEY": "${ZHIPU_API_KEY}"
      }
    }
  }
}

Exposed tools (3)

3 read · 0 write · 0 destructive.

Tool	Risk	Description
`site_search`	read	Search within a single website (restricted to one domain).
`web_fetch`	read	Fetch a URL and return its readable content as clean Markdown
`web_search`	read	Search the web using Baidu + Bing China. Returns title, URL, and snippet.

Trust audit

SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.

Layer	What it checks	Result
L0	Provenance & inventory	PASS
L1	Static analysis of the code	PASS
L2	Instruction surface (what it tells the agent)	PASS
L3	Class-specific surface	PASS
L4	Behavioural (sandbox)	SKIPPED

What the source does

Filesystem: none-observed
Network: declared (5 observation(s))
Shell: declared (2 observation(s))
Dependencies: not all pinned
Secrets in source: none-found