Skip to content

Available MCP Tools

There are two MCP servers with different tools:

  • Hosted MCP at https://app.crontinel.com/mcp. Connect with OAuth or a scoped key from /settings/ai-connections. This is the default.
  • Legacy MCP at https://app.crontinel.com/api/mcp, used by the @crontinel/mcp-server stdio package with an app API key. The tools act on the app the key belongs to.

Hosted MCP tools

Every hosted tool except get_connection takes a numeric app_id. The investigation tools appear only when hosted AI investigations are enabled.

ToolScopeDescription
get_connectionanyThe organization, app restrictions and scopes of this connection
list_monitorsmonitor:readMonitor status metadata for an app. Configurations and secrets are omitted
list_runsrun:readRecent runs, with outcome status and the AI quality grade when enabled
list_deploysrun:readDeploy markers (version, environment, commit SHA, time)
list_incidentsincident:readIncident start and resolution times
get_incident_evidenceall three read scopesA redacted evidence packet for one incident
get_fix_contextall three read scopesAn evidence packet for a coding agent
check_monitoring_coverageall three read scopesSaved monitors compared with the evidence actually received
explain_schedulemonitor:readValidate a saved cron monitor and preview up to five upcoming starts
start_investigationread scopes + investigation:runStart a hosted AI investigation. Spends one AI start
get_investigationall three read scopesStatus and evidence of an investigation this connection started
cancel_investigationread scopes + investigation:runCancel a queued or running investigation
save_cron_schedulemonitor:writeCreate or update the schedule monitor for one command
save_heartbeatmonitor:writeCreate or update a heartbeat monitor by name
save_outcome_rulemonitor:writeCreate or update the evidence a run must report to count as done

list_runs

Parameters:

  • app_id (required) – numeric app ID
  • limit (optional, 1 to 100) – runs per page
  • after_id (optional) – continue from next_after_id in the last response

Returns per run: status, exit code, duration, times, attempt, outcome_status, outcome_missing_keys, and quality. outcome_status is passed, failed, unknown, or null when no outcome rule applies. unknown means the receipt did not include the evidence named in outcome_missing_keys; it is not a success. quality is null unless the job has an AI quality check. When set, it holds status, score, pass_score, graded_at, and advisory: true. Output, command arguments, and the model’s reasons are not returned.

Configuration writes (hosted MCP)

The three save_* tools appear only when the connection has monitor:write. It is never granted by default: tick it when you create an API key, or let your assistant request it over OAuth, where the consent screen says what it allows.

  • The key or OAuth user needs operator access to the app. Viewers cannot write.
  • Each tool creates the record or updates the one with the same key (command for schedules, name for heartbeats, command plus metric key for outcome rules). Calling it twice does not make a duplicate.
  • Changes use the same checks, plan limits and audit log as the dashboard and the Configuration API. The audit entry shows via: mcp.
  • Ping URLs are not returned over MCP. Copy them from the dashboard.
  • Each save returns a version. Pass it back as if_version on the next save, and the save is refused if someone changed the record in between.
  • Deleting monitors, muting and maintenance windows are not MCP tools. Use the dashboard or the Configuration API.

Legacy MCP tools

ToolDescription
list_scheduled_jobsCommands that have reported runs, with their last run status
get_cron_statusLast run details for a command (exit code, duration, output)
get_queue_statusLaravel only. Depth, failed count, and oldest job age per queue
get_horizon_statusLaravel only. Horizon status, paused flag, supervisors, failed jobs per minute
list_recent_alertsAlerts fired in the last N hours
acknowledge_alertResolve an active alert by its key
create_alertCreate an alert channel for the app

list_scheduled_jobs

No parameters. Returns the commands that have reported runs for the key’s app, with their most recent status.

get_cron_status

Parameters:

  • command (required) – command name or partial match

Returns: command, status, exit code, duration, started_at, finished_at, output.

get_queue_status

Parameters:

  • queue (optional) – queue name. Returns all queues if omitted.

Returns: queue name, depth, failed count, oldest job age in seconds. Queue data comes from the Laravel package. An app that never reports it has no queue data, which is not the same as an empty queue.

get_horizon_status

No parameters. Returns: overall status, paused indicator, supervisors list, failed jobs per minute. Horizon is Laravel only. Missing data is not a stopped supervisor and not a healthy one.

list_recent_alerts

Parameters:

  • hours (optional, default 24) – look-back window

Returns: alert key, state (firing or resolved), fired_at, fire_count, resolved_at.

acknowledge_alert

Parameters:

  • alert_key (required) – the key from list_recent_alerts

Marks the alert as resolved. It is not the same as acknowledging in the dashboard: the alert closes, and it opens as a new alert if the condition fires again.

create_alert

Parameters:

  • type (required) – slack, email, webhook, discord, teams, telegram, pagerduty, sms, or opsgenie
  • One set of fields for the type, passed at the top level:
typerequired fields
slack, discord, teamswebhook_url
emailto
webhookurl
telegrambot_token and chat_id
pagerdutyrouting_key
smsphone_number (E.164, such as +15551234567)
opsgenieapi_key

pagerduty, sms, and opsgenie need a paid plan.

Example – create a Slack alert:

{
"type": "slack",
"webhook_url": "https://hooks.slack.com/services/T000/B000/xxxx"
}

Returns created, channel_id, and type on success.