# ── Zabbix connection ─────────────────────────────────────────────────────────

ZABBIX_URL=https://zabbix.example.com
ZABBIX_VERIFY_SSL=true

# ── Authentication ────────────────────────────────────────────────────────────

# API token (recommended — Zabbix 5.4+)
# Create one at: Administration → Users → your user → API tokens
ZABBIX_TOKEN=your_api_token_here

# User / password (alternative — remove ZABBIX_TOKEN if using this)
# ZABBIX_USER=Admin
# ZABBIX_PASSWORD=zabbix

# ── Agent: LLM provider ───────────────────────────────────────────────────────

# Active provider: nvidia | openrouter | groq  (default: nvidia)
ZABBIX_MCP_PROVIDER=nvidia

# NVIDIA Build (https://build.nvidia.com → get API key)
ZABBIX_MCP_NVIDIA_API_KEY=nvapi-...
ZABBIX_MCP_NVIDIA_MODEL=meta/llama-3.3-70b-instruct

# OpenRouter (alternative)
# ZABBIX_MCP_OPENROUTER_API_KEY=sk-or-...
# ZABBIX_MCP_OPENROUTER_MODEL=anthropic/claude-haiku-4-5

# Groq (alternative)
# ZABBIX_MCP_GROQ_API_KEY=gsk_...
# ZABBIX_MCP_GROQ_MODEL=llama-3.3-70b-versatile

# ── MCP transport ─────────────────────────────────────────────────────────────

# stdio  — JSON-RPC 2.0 over stdin/stdout (default, used by Claude Desktop and agents)
# http   — Streamable HTTP server on MCP_HOST:MCP_PORT (endpoint MCP_PATH, default /mcp)
# sse    — DEPRECATED alias of http; the legacy /sse + /messages/ endpoints are
#          served in HTTP mode anyway, for clients that cannot speak Streamable HTTP
MCP_TRANSPORT=stdio

# Bind address, port and endpoint path — only used in HTTP mode
# MCP_HOST=0.0.0.0
# MCP_PORT=8000
# MCP_PATH=/mcp

# Streamable HTTP tuning — all optional
# MCP_JSON_RESPONSE=false   # plain JSON replies instead of an SSE stream
# MCP_STATELESS=false       # fresh transport per request, no session tracking
# MCP_LEGACY_SSE=true       # set to false to drop the deprecated /sse endpoints