mcp-searxng MCP — 使用场景, 安装 & 实时演示

为什么要用

核心特性

Two tools: searxng_web_search and web_url_read
Search filters: pagination, time range, language, safe search
web_url_read returns clean Markdown with heading/paragraph filtering
Works with self-hosted or public SearXNG instances via SEARXNG_URL
HTTP transport via MCP_HTTP_PORT, plus STDIO default

实时演示

实际使用效果

searxng.replay ▶ 就绪

0/0

安装

选择你的客户端

~/Library/Application Support/Claude/claude_desktop_config.json · Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "searxng": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-searxng"
      ],
      "_inferred": true
    }
  }
}

打开 Claude Desktop → Settings → Developer → Edit Config。保存后重启应用。

~/.cursor/mcp.json · .cursor/mcp.json

{
  "mcpServers": {
    "searxng": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-searxng"
      ],
      "_inferred": true
    }
  }
}

Cursor 使用与 Claude Desktop 相同的 mcpServers 格式。项目级配置优先于全局。

VS Code → Cline → MCP Servers → Edit

{
  "mcpServers": {
    "searxng": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-searxng"
      ],
      "_inferred": true
    }
  }
}

点击 Cline 侧栏中的 MCP Servers 图标，然后选 "Edit Configuration"。

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "searxng": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-searxng"
      ],
      "_inferred": true
    }
  }
}

格式与 Claude Desktop 相同。重启 Windsurf 生效。

~/.continue/config.json

{
  "mcpServers": [
    {
      "name": "searxng",
      "command": "npx",
      "args": [
        "-y",
        "mcp-searxng"
      ]
    }
  ]
}

Continue 使用服务器对象数组，而非映射。

~/.config/zed/settings.json

{
  "context_servers": {
    "searxng": {
      "command": {
        "path": "npx",
        "args": [
          "-y",
          "mcp-searxng"
        ]
      }
    }
  }
}

加入 context_servers。Zed 保存后热重载。

claude mcp add searxng -- npx -y mcp-searxng

一行命令搞定。用 claude mcp list 验证，claude mcp remove 卸载。

使用场景

实战用法： mcp-searxng

Give your AI private web search via self-hosted SearXNG

👤 Privacy-minded devs, self-hosters ⏱ ~10 min intermediate

何时使用： You don't want Tavily/Brave/Google to see every AI query.

前置条件

A SearXNG instance — docker run -p 8080:8080 searxng/searxng (see docs for config)
SEARXNG_URL env var — Set to http://localhost:8080 or your remote instance

步骤

Search

Search: 'mcp protocol spec 2026'. Time range last 3 months.✓ 已复制

→ Result list
Read top result

web_url_read on the top result — give me the headings and first 5 paragraphs.✓ 已复制

→ Markdown excerpt
Follow up

Based on that page, what's the single most important change in the spec?✓ 已复制

→ Grounded answer

结果： Full research loop with zero third-party key exposure.

注意事项

SearXNG instance rate-limits or fails due to upstream engine blocking it — Rotate engines in SearXNG settings; consider running multiple instances
Self-hosted SearXNG needs periodic updates as engines change — Pin image tag and update monthly

搭配使用： filesystem

Fact-check a claim against multiple sources

👤 Writers, analysts ⏱ ~10 min beginner

何时使用： You need to confirm something across a few independent sources.

步骤

Search

Search the exact claim phrase. Return top 10 from different domains.✓ 已复制

→ Diverse result list
Cross-read

web_url_read on the 3 most authoritative — quote the sections that speak to the claim.✓ 已复制

→ Quoted excerpts
Verdict

Claim: supported, contested, or false? Cite specifically.✓ 已复制

→ Verdict with citations

结果： A cited fact-check.

Build a zero-API-key research assistant for your team

👤 Teams that won't/can't buy search API keys ⏱ ~40 min intermediate

何时使用： Everyone wants AI search but no one wants to manage Tavily keys per person.

步骤

Run SearXNG behind your intranet

Set up SearXNG container on the team server, point mcp-searxng at it.✓ 已复制

→ All team AI clients search via shared instance
Test from multiple clients

Run the same query from Claude Desktop and Cursor — both should hit the same SearXNG.✓ 已复制

→ Consistent results

结果： Team-wide AI search with one instance, no per-user billing.

组合

与其他 MCP 搭配，撬动十倍杠杆

searxng + filesystem

Research and save notes

Search 'SBOM best practices', web_url_read the top 3, save a distilled memo to ./notes/sbom.md.✓ 已复制

searxng + kindly-web-search

Stack both — kindly for platform-aware full fetches, searxng for private general search

Use searxng for the initial search; if any result is SO/GitHub, fall back to kindly for richer content.✓ 已复制

工具

此 MCP 暴露的能力

工具	输入参数	何时调用	成本
searxng_web_search	query: str, pageno?: int, time_range?: day\|week\|month\|year, language?: str, safesearch?: 0\|1\|2	Any web search	free (your SearXNG)
web_url_read	url: str, start_heading?: int, end_heading?: int, start_paragraph?: int, end_paragraph?: int	Clean extraction of an article	free

成本与限制

运行它的成本

API 配额: Unlimited (your SearXNG instance)
每次调用 Token 数: Search: 500-1500 tokens. Read: 1-10k depending on page.
费用: Free
提示: Narrow queries with time_range and language — cuts both tokens and SearXNG load.

安全

权限、密钥、影响范围

凭据存储： No keys needed; just SEARXNG_URL

数据出站： Your SearXNG instance (which in turn queries upstream engines)

Public SearXNG instances can be monitored by their operators — host your own for real privacy.
Extracted content reflects the live page — verify for critical decisions.

故障排查

常见错误与修复

Connection refused

SearXNG not running or SEARXNG_URL wrong. curl the URL directly first.

验证： curl $SEARXNG_URL

Empty results for any query

SearXNG blocked by upstream engines or misconfigured. Check /search directly in browser.

验证： Open $SEARXNG_URL/search?q=test

web_url_read returns mostly nav junk

Some sites are JS-heavy; narrow with heading/paragraph ranges, or use a different extractor.

替代方案

mcp-searxng 对比其他方案

替代方案	何时用它替代	权衡
omnisearch	You want multi-provider + AI search	Requires provider API keys
kindly-web-search	You want platform-aware full content	Still needs Serper/Tavily key unless paired with SearXNG
brave-search	You're OK with one commercial provider	Paid above the free tier

mcp-searxng