web-research-mcp
Uses Brave Search as an upstream engine for web search, allowing site-restricted queries and evidence-backed snippets.
Uses Google Search as an upstream engine for web search, with site filtering and page content retrieval.
Provides web search through a self-hosted SearXNG aggregator, supporting configurable engines, site filtering, and returning extractable evidence from result pages.
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-research-mcpSearch modelcontextprotocol.io for MCP server docs and extract evidence from the top result."
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 Research MCP
免费、免搜索 API Key 的本地联网 MCP 服务。提供 websearch 和 webfetch:按域名搜索、读取网页、返回原文证据、相关性说明与可追溯的快照。
上游使用自建 SearXNG 的匿名网页适配器。无需搜索账号、免费 Key 或商业试用额度,不自动回退到付费服务。本项目采用 MIT,SearXNG 是独立部署的 AGPL-3.0-or-later 项目。机器和网络仍有成本,免费上游可能限流、验证码或零结果,服务会明确报告。上游准入规则
已实现
指定站点:sites、exclude_domains、子域开关、IDNA/PSL 校验;返回结果和证据跳转都检查范围。
原文证据:显式 evidence_mode=extract 有界补抓;返回连续 quote、来源、完整文本哈希、Unicode 偏移和 snapshot cursor。
可解释评分:词法相关性记录方法、版本与依据;confidence 表示证据可追溯性,fact_probability 固定为 null。
持久化续读:SQLite 保存冻结候选池、网页快照与随机游标,重启后在有效期内仍可读取。
受控抓取:DNS 固定连接、SSRF/跳转检查、robots、压缩/输出上限、取消、独立解析 worker 与资源清理。
真实 MCP 入口:SDK 2、stdio 和旧版初始化兼容路径;结构化输出与文本输出一致。
版本为 0.1.0,业务契约为 0.2-draft。默认评分是词法基线;PDF、动态浏览器、向量召回、BM25/RRF 实验和神经精排不属于本版本。high 证据等级不表示事实一定正确。
Related MCP server: qsearch
快速开始
需要 Node 24.20.0、pnpm 10.12.3 和 Docker Compose。Node 版本记录在 .node-version。
git clone https://github.com/wyzz973/web-research-mcp.git
cd web-research-mcp
pnpm install --frozen-lockfile
pnpm searxng:setup
docker compose -f deploy/compose.yaml up -d
pnpm buildSearXNG 只监听 127.0.0.1:18888。初始化将随机实例 secret 写入被 Git 忽略的 .cache/searxng/settings.yml;这不是搜索 API Key。固定镜像和精确引擎配置见 部署说明。
运行一次带原文证据的搜索:
pnpm call websearch '{"query":"MCP tools structuredContent","sites":["modelcontextprotocol.io"],"limit":3,"evidence_mode":"extract","max_evidence_results":2}' --config config/local.example.json读取公开网页:
pnpm call webfetch '{"url":"https://www.sqlite.org/fts5.html","format":"text","max_chars":3000}'把搜索返回的 evidence[].snapshot_cursor 或抓取返回的 next_cursor 原样传入,即可读取已保存文本。以下占位值要替换为实际响应:
pnpm call webfetch '{"cursor":"ACTUAL_CURSOR","max_chars":3000}'示例使用匿名 Brave 网页搜索 + Google 网页搜索,不使用商业 Search API。可选择其他经准入的引擎;可达性与结果质量会变化。
连接 MCP 客户端
配置客户端使用 Node 24 启动 dist/mcp/stdio.js。以下为常见 mcpServers 格式,路径需替换成本机路径:
{
"mcpServers": {
"web-research": {
"command": "/absolute/path/to/node24/bin/node",
"args": ["/absolute/path/to/web-research-mcp/dist/mcp/stdio.js"],
"env": {
"SEARXNG_URL": "http://127.0.0.1:18888",
"SEARXNG_ENGINES": "brave,google"
}
}
}
}也可用 --config file.json 指定配置。WEB_RESEARCH_DATA_DIR 覆盖数据目录;默认在用户目录 .local/share/web-research-mcp。搜索未配置时 webfetch 仍可用,非法配置启动失败。
stdio 会等待客户端输入;没有普通命令行输出是正常行为。stdout 只传 MCP,日志写 stderr。--help 和 --version 为独立 CLI 查询。
开发与验收
pnpm check
pnpm test:built
SEARXNG_URL=http://127.0.0.1:18888 SEARXNG_ENGINES=brave,google pnpm test:livecheck 包括类型、类型感知 lint、格式、生成类型、Schema、模块边界、文档、离线行为与 MCP 测试。test:built 将真实 tarball 安装到干净目录,验证 SQLite、MCP 入口和解析 worker。test:live 明确访问上游,不放入普通 CI;失败会保存到被 Git 忽略的 artifacts/live/result.json,不自动绕过封锁。
执行证据见 实施验收记录。离线测试不能保证所有网站随时可抓取。
文档
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections. Dates show when Glama detected each change.
No tool schema history has been recorded yet.
This server cannot be installed
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Related MCP Connectors
Stealth web browser for agents: search, fetch, click, download and type in persistent MCP sessions.
Live AI-native web search with citations. One tool for every MCP client. Flat per-request pricing.
One MCP for the Web. Easily search, crawl, navigate, and extract websites without getting blocked.…
Verified business OSS MCP for search, RSS, crawling, documents, browser, media and transcription.
Related MCP Servers
- AlicenseAqualityBmaintenanceProvides local-first web intelligence over MCP with tools for search, fetch, crawl, extract, cache, find-similar, research, and autonomous agent loops, requiring no API keys.107635,202AGPL 3.0
- AlicenseNot gradedqualityBmaintenanceEnables AI agents to perform web searches with full content retrieval and multi-engine provenance, including trust scoring and local corpus persistence, via MCP integration.32Apache 2.0
- AlicenseNot gradedqualityBmaintenanceMCP server enabling local-first web search, fetch, extract, and caching with citeable excerpts, no API key required. Supports research workflows for agents and apps.18MIT
- AlicenseNot gradedqualityBmaintenanceEnables AI agents to perform local-first web search, fetch, crawl, extract, cache, research, and autonomous information gathering through MCP, with no API keys or cloud dependencies.763AGPL 3.0
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/wyzz973/web-research-mcp'
If you have feedback or need assistance with the MCP directory API, please join our Discord server