MCPIndex — Camada de Confiança para Servidores MCP
mcpindex-ai/mcpindex-web
Veredictos de confiança para cada ferramenta MCP antes da execução
Ver no GitHubÍndice agent-nativo que fornece vereditos de confiança (ALLOW/DENY/REVIEW) sobre servidores MCP antes que seu agente execute ferramentas. Oferece API de confiança por ferramenta, MCP server pronto para instalar e superfícies legíveis por agentes (diretório indexado, llms.txt, JSON-LD).
Instalação
npx convez-ai mcpindex-ai/mcpindex-web1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Validar segurança e confiabilidade de servidores MCP antes de agentes os executarem
- Descobrir e avaliar qualidade de 3.500+ servidores MCP catalogados
- Integrar vereditos de confiança em pipelines de agentes Claude/Cursor/Cline
Veredito

Ferramental crítico para ecossistema MCP — transforma descoberta de ferramentas em processo confiável com metodologia aberta. API versioned, MCP server distribuído via npm (MIT), e infraestrutura Next.js bem documentada tornam isso production-ready para agents precisarem validar terceiros.
README
A tradução para o português deste README ainda está na fila.
mcpindex.ai
The trust layer for MCP tools.
A verdict (ALLOW / DENY / REVIEW) on whether an MCP tool does what it claims, before your agent acts. Your agent trusts every tool it is handed; mcpindex doesn't.
Live site · Trust · Methodology · Brand · npm: mcp-server-mcpindex
Three primitives
- Trust verdict API -
GET /api/v1/trust/tool/<server_id>/<tool_name>returns a per-tool verdict: decision + dimensions + severity. Fail-closed; advisory at v1. - Drop-in MCP server -
npm install -g mcp-server-mcpindex. Add to Claude Desktop / Cursor / Cline / Zed; askcheck_tool_trustbefore invoking a tool it just discovered. - Agent-readable surfaces - indexed directory plus
/llms.txt,/.well-known/mcp-index.json, and JSON-LD on every page.
Develop
npm install
node scripts/fetch-snapshot.mjs # one-time: pull current registry snapshot
npm run dev # http://localhost:3000
Snapshot lives at data/snapshot.json (committed). Refresh anytime with the script.
Stack
- Next.js 16 (App Router) on Vercel
- Tailwind v4
- Local snapshot of
registry.modelcontextprotocol.io/v0/servers, refreshed daily via Vercel cron - Quality Score:
lib/quality.ts(open methodology - PRs welcome) - Search:
lib/search.ts(keyword now; embeddings v2 when OPENAI_API_KEY is wired)
Project layout
mcpindex/
├── app/ # Next.js routes
│ ├── api/v1/ # Versioned public API
│ ├── api/cron/ # Vercel-cron-driven endpoints
│ ├── server/[slug]/ # 3,500+ per-server pages (generateStaticParams)
│ ├── best/[category]/ # 28 curated category pages
│ └── ... # leaderboard, changelog, methodology, about, pricing, terms, privacy, stats
├── components/ # Header, Footer, AgentDemo (live demo), LiveTicker
├── lib/ # registry, quality, search, categorize, installs, types
├── scripts/ # sync-registry.mjs, fetch-snapshot.mjs (one-shot)
├── data/ # snapshot.json (committed) + snapshots/ (cron-written)
├── launch/ # all launch-day copy: Show HN, Reddit, cold emails, pitches
├── mcp-server-mcpindex/ # the npm-distributed MCP server
└── READY_TO_LAUNCH.md # ops checklist for going live
License
- Web app code: source-available, all-rights-reserved (prevents direct fork-and-deploy by competitors).
mcp-server-mcpindex(the npm package): MIT.- MCP Quality Score methodology: open and PR-friendly.
Affiliation
Unofficial. Not affiliated with Anthropic. The Model Context Protocol is open under MIT and trademarks remain with their owners. Server data comes from the official MCP registry.