Skip to main content

​
Usage

langsight monitor [OPTIONS]

​
Options

OptionDefaultDescription
--configauto-discoveredPath to .langsight.yaml
--interval, -i30Health check interval in seconds
--oncefalseRun one cycle and exit (useful for cron/CI)

​
What it does

Polls all configured servers on a configurable interval. After each cycle, results are evaluated by the alert engine — alerts fire on state transitions, not on every check. Alert types:
  • SERVER_DOWN — after N consecutive failures (configurable)
  • SERVER_RECOVERED — when a DOWN server comes back
  • SCHEMA_DRIFT — when tool schema changes unexpectedly
  • HIGH_LATENCY — when latency exceeds 3× baseline
Deduplication: A single incident generates one alert, not one per polling cycle.

​
Example

langsight monitor --interval 30

Cycle #1 — next in 30s
──────────────────────────────────────────────
Server           Status      Latency   Tools
postgres-mcp     ✓ up        142ms     5
s3-mcp           ✓ up        890ms     7

[Alert] CRITICAL — MCP server 'jira-mcp' is DOWN
  Server has been unreachable for 2 consecutive checks.
  Slack notification sent.

​
Alert configuration

Configure in .langsight.yaml: 
alerts:
  slack_webhook: https://hooks.slack.com/services/...
  consecutive_failures: 2        # checks before DOWN alert
  latency_spike_multiplier: 3.0  # 3x baseline = HIGH_LATENCY
  error_rate_threshold: 0.05     # 5% error rate threshold

​
Cron / one-shot usage

# Run once (e.g. every 5 minutes via cron)
langsight monitor --once

# Crontab entry
*/5 * * * * /usr/local/bin/langsight monitor --once >> /var/log/langsight.log 2>&1