kindly-web-search-mcp-server MCP — Cas d'usage, Installer & Démo en direct

Pourquoi l'utiliser

Fonctionnalités clés

web_search returns page_content as Markdown, not just title/snippet
get_content does platform-aware full fetches (SO threads, GitHub Issues, arXiv)
Supports Serper (primary), Tavily (fallback), or self-hosted SearXNG
Optional GITHUB_TOKEN enables structured GitHub Issue extraction
Works from Claude Code, Codex, Cursor, Copilot, Gemini, Antigravity

Démo en direct

Aperçu en pratique

kindly-web-search.replay ▶ prêt

0/0

Installer

Choisissez votre client

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

{
  "mcpServers": {
    "kindly-web-search": {
      "command": "uvx",
      "args": [
        "kindly-web-search-mcp-server"
      ],
      "_inferred": true
    }
  }
}

Ouvrez Claude Desktop → Settings → Developer → Edit Config. Redémarrez après avoir enregistré.

~/.cursor/mcp.json · .cursor/mcp.json

{
  "mcpServers": {
    "kindly-web-search": {
      "command": "uvx",
      "args": [
        "kindly-web-search-mcp-server"
      ],
      "_inferred": true
    }
  }
}

Cursor utilise le même schéma mcpServers que Claude Desktop. La config projet l'emporte sur la globale.

VS Code → Cline → MCP Servers → Edit

{
  "mcpServers": {
    "kindly-web-search": {
      "command": "uvx",
      "args": [
        "kindly-web-search-mcp-server"
      ],
      "_inferred": true
    }
  }
}

Cliquez sur l'icône MCP Servers dans la barre latérale Cline, puis "Edit Configuration".

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "kindly-web-search": {
      "command": "uvx",
      "args": [
        "kindly-web-search-mcp-server"
      ],
      "_inferred": true
    }
  }
}

Même format que Claude Desktop. Redémarrez Windsurf pour appliquer.

~/.continue/config.json

{
  "mcpServers": [
    {
      "name": "kindly-web-search",
      "command": "uvx",
      "args": [
        "kindly-web-search-mcp-server"
      ]
    }
  ]
}

Continue utilise un tableau d'objets serveur plutôt qu'une map.

~/.config/zed/settings.json

{
  "context_servers": {
    "kindly-web-search": {
      "command": {
        "path": "uvx",
        "args": [
          "kindly-web-search-mcp-server"
        ]
      }
    }
  }
}

Ajoutez dans context_servers. Zed recharge à chaud à la sauvegarde.

claude mcp add kindly-web-search -- uvx kindly-web-search-mcp-server

Une seule ligne. Vérifiez avec claude mcp list. Supprimez avec claude mcp remove.

Cas d'usage

Usages concrets : kindly-web-search-mcp-server

Debug a library error by reading the whole StackOverflow thread

👤 Developers hit with a cryptic error ⏱ ~10 min beginner

Quand l'utiliser : The top SO answer is often not right; the conversation below has the real fix.

Prérequis

SERPER_API_KEY — serper.dev — 2500 free queries/mo

Déroulement

Search the exact error

Search for: 'fatal error: 'openssl/opensslv.h' file not found' arm64 mac — return the top 3 SO URLs.✓ Copié

→ SO URLs with snippets
Fetch the full thread of the most relevant

get_content on the top SO URL — give me the whole thread including every comment.✓ Copié

→ Full Q+A+comments in Markdown
Pick the fix the thread's consensus supports

Which answer (including ones below the accepted one) does the thread's comments actually validate? Show me the exact commands.✓ Copié

→ Specific command sequence

Résultat : The real fix, not just the accepted answer — with evidence from comments.

Pièges

SO pages can be huge; full content can be 10k+ tokens — Ask for an executive summary first, then drill in

Combiner avec : filesystem

Read a long GitHub Issue with full context

👤 Engineers evaluating whether to upgrade a dependency ⏱ ~10 min beginner

Quand l'utiliser : The issue has 80 comments and you need the signal from the noise.

Prérequis

GITHUB_TOKEN (optional but recommended) — A classic/fine-grained token; empty scope set fine for public repos

Déroulement

Pull structured issue data

get_content on https://github.com/vercel/next.js/issues/12345 — return structured metadata plus body and all comments.✓ Copié

→ Structured extract, not raw HTML
Distill signal

Summarize: is this fixed, open, workaround-only? Who from the maintainer team has commented, and what was their last position?✓ Copié

→ Status + maintainer stance

Résultat : A grounded go/no-go decision on a library upgrade.

Combiner avec : github

Summarize an arXiv paper with full text access

👤 Researchers, curious devs ⏱ ~10 min beginner

Quand l'utiliser : You don't want a snippet; you want to quote section 3.2.

Déroulement

Fetch the paper

get_content https://arxiv.org/abs/2410.XXXXX — full text.✓ Copié

→ Markdown of the paper
Targeted summary

Summarize the methodology (section 3) in 5 bullets and quote the key claim from the results table.✓ Copié

→ Summary with verbatim quote

Résultat : Paper understanding without leaving chat.

Combinaisons

Associez-le à d'autres MCPs pour un effet X10

kindly-web-search + github

Kindly reads the issue discussion; github MCP creates a tracking issue in your own repo

Read the discussion on open-upstream/project#432 via kindly, then open a shadowing issue in our team's repo summarizing it.✓ Copié

kindly-web-search + filesystem

Save researched answers as Markdown notes

Research 'best practices for Postgres connection pooling with pgbouncer', save the full SO thread + two docs links under ./notes/pgbouncer.md.✓ Copié

Outils

Ce que ce MCP expose

Outil	Entrées	Quand appeler	Coût
web_search	query: str, num_results?: int (default 3)	Any web search where you'd have followed with a fetch anyway	1 search provider call + N fetches
get_content	url: str	Deep dive on a specific URL	1 fetch (plus provider ops for platform parsing)

Coût et limites

Coût d'exécution

Quota d'API: Serper: 2500/mo free. Tavily: 1000/mo free. SearXNG self-hosted: unlimited.
Tokens par appel: web_search with full content: 2-6k per call. get_content: 1-10k depending on page.
Monétaire: Free with provider free tiers; self-host SearXNG for zero cost
Astuce: For iterative research, lower num_results to 2 — kindly returns full content per result, so 3 can be expensive.

Sécurité

Permissions, secrets, portée

Stockage des identifiants : Env vars: SERPER_API_KEY or TAVILY_API_KEY or SEARXNG_BASE_URL

Sortie de données : Queries to chosen search provider + fetches to result URLs

Ne jamais accorder : GitHub token with repo write — public-read is enough for content extraction

Returned content includes whatever the live page serves — verify before trusting for security-critical decisions.
Requires a Chromium browser on the host for full-content rendering.

Dépannage

Erreurs courantes et correctifs

No provider configured

Set at least one of SERPER_API_KEY, TAVILY_API_KEY, or SEARXNG_BASE_URL.

Vérifier : env | grep -E 'SERPER|TAVILY|SEARXNG'

Chromium not found

Install Chrome, Edge, Brave, or chromium. Set PUPPETEER_EXECUTABLE_PATH if auto-detect fails.

Vérifier : which google-chrome || which chromium

Python version error

Requires Python 3.13+. Upgrade via pyenv or uv.

Vérifier : python3 --version

Alternatives

kindly-web-search-mcp-server vs autres

Alternative	Quand l'utiliser	Compromis
omnisearch	You want many providers including AI-search in one MCP	Less platform-aware; snippets-first
brave-search	You only want search, not full-content retrieval	Snippets only
fetch MCP	You don't need search, just URL fetching	No search

Plus

Ressources

📖 Lire le README officiel sur GitHub

🐙 Voir les issues ouvertes

🔍 Parcourir les 400+ serveurs MCP et Skills