Levirge Search
Web search and page reading for AI agents, including JSON APIs, RSS feeds and YouTube.
Installation
Ajoutez ceci à la configuration MCP de votre client (Claude Desktop : claude_desktop_config.json ; Cursor : .cursor/mcp.json ; VS Code et les autres utilisent la même forme « mcpServers »).
claude mcp add --transport http levirge-search 'https://search.levirge.com/mcp'
{
"mcpServers": {
"levirge-search": {
"type": "http",
"url": "https://search.levirge.com/mcp"
}
}
}
Documentation
Tirée du README du projet sur GitHub, actualisée chaque semaine. Les liens ouvrent GitHub.
Sommaire
Levirge Search — web research your agent can actually reach
Search the web and read the sources, with citations.
Levirge Search is a hosted MCP server for research: unfiltered web search, and a URL fetch where pages that need JavaScript or refuse plain requests are opened in a full browser — so your agent can open the original source and cite it.
Install (2 minutes)
Claude Code:
claude plugin marketplace add levirge/search
claude plugin install levirge-search@levirge-search
Then run /mcp in a session and sign in.
Codex CLI:
codex plugin add levirge/search
Any other MCP client — point it at https://search.levirge.com/mcp. It answers a standard OAuth
challenge, so clients that speak MCP authorization discover sign-in on their own; no token to paste.
Update
plugin install won't upgrade, and the desktop "Update" button stays greyed out until the app's
marketplace cache is refreshed (the dialog has no refresh button). From the CLI — it shares the same
store as the desktop app:
claude plugin marketplace update levirge-search
claude plugin update levirge-search@levirge-search
Then restart the client — the plugin registry and MCP tool schemas are cached at startup, so new commands and tool params don't appear until then.
What you get
- Search that isn't a summary —
searchruns SearXNG across engines and returns real result sets, not one provider's ranking. - Fetch that opens the page —
fetchtakes any URL, including agent-constructed API endpoints; pages that need JavaScript or refuse plain requests are opened in a full browser. One tool, both routes. - Background scans —
fetch_asyncqueues up to 200 URLs through the same routing and paces them per-domain; collect withfetch_resultsinstead of blocking. - A stealth registry that learns — domains that needed the full browser are remembered per
tenant, so the second visit skips the retry (
/levirge-search:stealth). - A failure queue worth reading —
stealth_failureslists domains where the stealth browser itself got blocked, including regressions on domains that used to work. - A skill that fires on refusal —
web-fetch-fallbackteaches the agent to reach for this when its primary fetch says "blocked", instead of silently falling back to search.
Commands
| Command | Does |
|---|---|
/levirge-search:research | Multi-query search, fetch the top hits, synthesize with sources |
/levirge-search:fetch | Fetch one URL via routed fetch and summarize it |
/levirge-search:stealth | List stealth-required domains, or register one |
Links
- App / sign in: https://search.levirge.com
- Levirge: https://levirge.com
- Plugin reference: plugins/levirge-search/README.md
Questions fréquentes
Qu'est-ce que Levirge Search ?
Web search and page reading for AI agents, including JSON APIs, RSS feeds and YouTube.
Comment installer Levirge Search ?
Il fonctionne comme serveur distant. Dans Claude Code : claude mcp add --transport http levirge-search 'https://search.levirge.com/mcp'. Pour les autres clients, collez la configuration JSON de cette page dans leurs réglages MCP.
Faut-il une clé API pour Levirge Search ?
Son entrée au registre n'indique aucune clé ni secret obligatoire. Certaines fonctions peuvent tout de même demander un compte auprès du service auquel il se connecte.
Levirge Search est-il open source ?
Son code est public sur GitHub, mais le dépôt ne déclare pas de licence standard.
Qui maintient Levirge Search ?
Il est publié par levirge. AI.tn le référence sans le maintenir ni le recommander.
Issu du registre MCP officiel. Être listé n'est pas une recommandation : vérifiez le code et les permissions d'un serveur avant de lui donner accès à vos données.