claude-plugins/infraestrutura/skills/gateway-check/SKILL.md

name, description, context

name	description	context
gateway-check	Gestao completa dos MCPs no gateway.descomplicar.pt — health check, restart, troubleshoot, mapa de portas, adicionar MCPs. Usar quando MCP falha, health check, ou gestao gateway.	fork

/gateway-check v2.0

Gestao e health check dos MCPs no servidor gateway (VM 103, gateway.descomplicar.pt).

Referencia: Memory mcp-gateway.md | PROC-MCP-Desenvolvimento.md

---

Acesso

• VM: 103 no Proxmox (QEMU)
• IP: 5.9.90.69
• SSH: mcp__ssh-unified__ssh_execute(server="gateway")
• HTTPS: https://gateway.descomplicar.pt/v1/<nome>/mcp
• Nginx whitelist: 188.251.199.30 (IP fixo NOS). Se 403 -> verificar IP com curl -4 ifconfig.me
• Nginx config: /etc/nginx/sites-enabled/ no gateway
• NAO confundir com: server (VM 100, 5.9.90.105), easy (VM 101, 5.9.90.70), dev (LXC 102)

---

Mapa de MCPs (30 services — actualizado 28-03-2026)

pm2 (Node.js — /opt/mcp-gateway/)

pm2 ID	Nome	Porta	nginx path	Prioridade
0	mcp-desk-crm	3150	/v1/desk-crm/mcp	P1
1	mcp-memory	3151	/v1/memory/mcp	P1
2	mcp-wikijs	3152	/v1/wikijs/mcp	P3
5	mcp-moloni	3158	/v1/moloni/mcp	P2
6	mcp-youtube-research	3157	/v1/youtube-research/mcp	P3
7	mcp-youtube	3187	/v1/youtube/mcp	P3

systemd (24 services)

Service	Porta	nginx path	Tipo	Prioridade
google-workspace-mcp	3164	/v1/google-workspace/mcp	Python/FastMCP	P1
mcp-time	3163	/v1/mcp-time/mcp	Node	P1
imap-enterprise	3160	/v1/imap/mcp	Node	P2
gitea-mcp	3162	/v1/gitea/mcp	Go	P2
n8n-mcp	3161	/v1/n8n/mcp	Node	P2
gsc-mcp	3153	/v1/gsc/mcp	Python/FastMCP	P2
google-analytics-mcp	3156	/v1/google-analytics/mcp	Python/FastMCP	P2
context7-mcp	3159	/v1/context7/mcp	Node	P2
mcp-reonic	3170	/v1/reonic/mcp	Node	P3
cloudflare-dns-mcp	3171	/v1/cloudflare-dns/mcp	supergateway	P3
magic-mcp	3172	/v1/magic/mcp	supergateway	P3
mcp-echarts-mcp	3173	/v1/mcp-echarts/mcp	supergateway	P3
mcp-mermaid-mcp	3174	/v1/mcp-mermaid/mcp	supergateway	P3
metabase-mcp	3175	/v1/metabase/mcp	supergateway	P3
pixabay-mcp	3176	/v1/pixabay/mcp	supergateway	P3
replicate-mcp	3177	/v1/replicate/mcp	supergateway	P3
outline-api-mcp	3178	/v1/outline-api/mcp	supergateway	P3
pexels-mcp	3179	/v1/pexels/mcp	supergateway	P3
penpot-mcp	3180	/v1/penpot/mcp	supergateway	P3
vimeo-mcp	3181	/v1/vimeo/mcp	supergateway	P3
presenton-mcp	3182	/v1/presenton/mcp	supergateway	P3
cwp-mcp	3183	/v1/cwp/mcp	supergateway	P3
design-engine-mcp	3184	/v1/design-engine/mcp	supergateway	P3

Prioridades: P1=critico (bloqueia trabalho) | P2=importante (degrada workflow) | P3=util Proxima porta livre: 3188

---

Protocolo de Health Check

Executar via mcp__ssh-unified__ssh_execute(server="gateway") em 2 chamadas:

Chamada 1 — estado geral

echo "=== PM2 ===" && pm2 list 2>/dev/null && echo "=== SYSTEMD ===" && systemctl list-units --type=service --state=running --no-pager | grep -i mcp && echo "=== FAILED ===" && systemctl list-units --type=service --state=failed --no-pager | grep -i mcp && echo "=== LOAD ===" && uptime && free -h

Chamada 2 — portas + erros

echo "=== PORTAS ===" && for port in 3150 3151 3152 3153 3156 3157 3158 3159 3160 3161 3162 3163 3164 3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3187; do if ss -tln | grep -q ":${port} "; then echo "OK   :${port}"; else echo "DOWN :${port}"; fi; done && echo "=== RAM MCPs ===" && ps aux | grep -E 'mcp|supergateway' | grep -v grep | awk '{sum+=$6} END {printf "Total: %.0f MB\n", sum/1024}' && echo "=== ERROS (30min) ===" && for svc in $(systemctl list-units --type=service --state=running | grep -i mcp | awk '{print $1}'); do errs=$(journalctl -u $svc --since "30 min ago" -p err --no-pager -q 2>/dev/null | wc -l); if [ "$errs" -gt 0 ]; then echo "--- $svc ($errs) ---"; journalctl -u $svc --since "30 min ago" -p err --no-pager -q 2>/dev/null | tail -2; fi; done && echo "=== PM2 ERROS ===" && pm2 logs --err --lines 3 --nostream 2>/dev/null

Output esperado

Apresentar como tabela resumo com data via mcp-time:

## Gateway Health — [data]
Servidor: gateway.descomplicar.pt | Load: X.XX | RAM: X.XG/XG | MCPs RAM: XXXMB
X/30 operacionais | Alertas: N

Criterios: OK=running+porta escuta | WARN=running com erros ou >5 restarts | DOWN=parado ou porta fechada

---

Operacoes

Restart de um MCP

# pm2
pm2 restart <nome>

# systemd
systemctl restart <nome>.service

Ver logs de um MCP

# pm2
pm2 logs <nome> --lines 30 --nostream

# systemd
journalctl -u <nome>.service --since "1h ago" --no-pager | tail -30

Testar endpoint especifico

# Internamente no gateway
curl -s http://127.0.0.1:<porta>/mcp -X POST \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"test","version":"0.1"}}}'

# Externamente
curl -s https://gateway.descomplicar.pt/v1/<nome>/mcp -X POST \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"test","version":"0.1"}}}'

---

Troubleshooting

MCP DOWN

1. Verificar service: systemctl status <nome> ou pm2 show <nome>
2. Ver logs: journalctl -u <nome> --since "1h ago" --no-pager | tail -30
3. Se supergateway: verificar preload catch-errors.mjs (ver abaixo)
4. Tentar restart: systemctl restart <nome>
5. Re-verificar porta: ss -tln | grep :<porta>

Supergateway crash (bug conhecido)

• Erro: No connection established for request ID: 0
• Fix: preload script em /opt/mcp-gateway/supergateway-catch-errors.mjs
• Activacao: Environment="NODE_OPTIONS=--import /opt/mcp-gateway/supergateway-catch-errors.mjs" no unit file
• 15 services patchados: todos os supergateway na tabela acima
• Ao adicionar novo supergateway: OBRIGATORIO adicionar esta linha ao unit file

FastMCP Python + nginx (DNS rebinding)

FastMCP 1.26+ bloqueia Host headers externos. No nginx:

proxy_set_header Host "127.0.0.1:<PORTA>";  # CORRECTO
# proxy_set_header Host $host;              # ERRADO — FastMCP bloqueia

Detectar: Se MCP retorna Invalid Host header via HTTPS mas funciona em curl localhost -> e este problema.

403 Forbidden

IP nao esta na whitelist nginx. Verificar: curl -4 ifconfig.me IP autorizado: 188.251.199.30. Actualizar em /etc/nginx/sites-enabled/ se mudou.

RAM total > 2GB

1. Identificar top consumers: ps aux --sort=-%mem | head -20 | grep -E 'node|python|supergateway'
2. Processos orphan: ps aux | grep -c supergateway
3. Se orphans > 24: pkill -f supergateway e restart escalonado

---

Adicionar novo MCP ao gateway

1. Instalar em /opt/mcp-gateway/<nome>/ (Node) ou /opt/mcp-<nome>/ (Python)
2. Porta: proxima livre (actualmente 3188)
3. Criar unit file systemd (se supergateway: incluir preload catch-errors)
4. Criar bloco nginx (se FastMCP Python: Host header fix obrigatorio)
5. systemctl daemon-reload && systemctl enable --now <nome>.service
6. nginx -t && systemctl reload nginx
7. Testar internamente e externamente (ver comandos curl acima)
8. Adicionar a ~/.claude.json: {"type":"http","url":"https://gateway.descomplicar.pt/v1/<nome>/mcp"}
9. Actualizar esta skill (mapa de portas + proxima porta livre)
10. Actualizar memory mcp-gateway.md

---

Anti-Patterns

• Nunca restart massivo sem verificar primeiro
• Nunca ignorar MCP P1 em estado DOWN
• Nunca confundir gateway (VM 103) com dev/server/easy
• Sempre reportar estado mesmo que tudo OK
• Sempre testar endpoint apos restart
• Sempre actualizar mapa de portas ao adicionar/remover MCPs

---

Skill v2.0.0 | 28-03-2026 | Descomplicar

---

Healing Log

Registo de erros conhecidos e como evitá-los. Lido automaticamente antes de executar.

{"date":"","issue":"","fix":"","source":"user|auto"}

Adicionar nova linha após cada erro corrigido.