convez.ai
MCP ServerIntegração & APIsIntermediário
ardp-project

ARD Connectors — Descoberta de Recursos Agênticos

ardp-project/connectors

Descubra e integre recursos agênticos via Agent Finder em múltiplas plataformas

Ver no GitHub
InicialPluginScore · de 0 a 100

Conectores cliente que permitem chatbots descobrir ferramentas, skills, MCP servers e APIs consultando serviços ARD/Agent Finder. Funciona com Claude, ChatGPT, Copilot e Gemini, oferecendo controle total sobre quais endpoints consultar e nunca instalando automaticamente.

Instalação

$npx convez-ai ardp-project/connectors

1 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

Estrelas
18
Forks
16
Acompanhando
0
Saúde
37%
Último commit
há 4 meses
Versão
—
Issues abertas
6
Contribuidores
1

Casos de uso

  • Localizar MCP servers e skills relevantes para uma tarefa específica
  • Integrar descoberta de recursos em chatbots multi-plataforma
  • Controlar quais endpoints de descoberta consultar antes de instalar qualquer ferramenta

Veredito

Solução técnica sólida para descoberta federada de recursos agênticos. O design client-side com consentimento explícito é adequado para ambientes corporativos, mas depende de endpoints ARD configurados manualmente — não funciona pronto para uso.

README

A tradução para o português deste README ainda está na fila.

ARD connectors

Client-side connectors that let a chatbot discover agentic resources — tools, skills, MCP servers, agents, and APIs — by searching ARD discovery services such as Agent Finder.

These connectors are client-side only. The discovery service (e.g. Agent Finder) exposes the HTTP POST /search and the remote MCP interface; this repo contains no server. Each connector just points your chatbot at an Agent Finder endpoint you choose.

The contract — what every connector does

Every connector follows the same flow (canonical text: shared/discovery-instructions.md). Nothing is installed automatically:

  1. Ask first — when you ask it to find tools/skills/agents for a task, it asks which Agent Finder endpoint(s) to query.
  2. Query those endpoints (POST /search).
  3. Present the ranked results for you to review (relevance score is not a trust or safety rating).
  4. Never auto-installs anything.
  5. Installs only on request — once you pick a result, it shows you how to add that resource yourself.

Quick install — Claude Code

This repo is a Claude Code plugin marketplace. Install the skill with:

/plugin marketplace add ards-project/connectors
/plugin install find-agentic-resources@ard-connectors

Custom skills are currently Claude Code–only (not the claude.ai web app or Desktop). For other platforms, see the per-platform folders below.

Two ways to connect

  • Skill — a portable instruction bundle that drives the flow over HTTP. Works on platforms with a "Skills"-style mechanism and an HTTP/Action capability.
  • MCP — add an Agent Finder remote MCP connector so the chatbot gets a native search tool, paired with the Skill for the behavior.

Per-platform connectors

PlatformSkillMCP
Claudeskills/find-agentic-resources/mcp/claude/
ChatGPTskills/chatgpt/mcp/chatgpt/
Microsoft Copilotskills/copilot/mcp/copilot/
Geminiskills/gemini/mcp/gemini/

Capability notes: ChatGPT Skills and Developer-mode MCP are gated to certain plans; the consumer Gemini app has neither custom MCP nor HTTP, so use the Gemini CLI or API there. Each folder spells out the specifics.

Endpoints

There is no built-in default Agent Finder. You choose which discovery services to trust by editing agent-finders.json — a list of endpoints the connector presents and asks you to pick from. The shipped entries (*.example.org, *.internal.example) are placeholders; replace them with real discovery services. The connector never queries an endpoint you didn't choose, and never auto-installs what it finds.

Ver o original no GitHub

Bom para

Relacionados