Integrations

MCP Tools Reference

MCP Tools Reference

These are the tools an AI assistant can call once connected to Sentinel's MCP server. They reach every team you own or belong to (see Teams below). You don't call them directly; your assistant picks the right tool from your natural-language request. This reference documents what each one accepts and returns. Every tool is available on every plan, the free plan included (except check_url, which needs a paid plan); the write tools at the end need a verified email on the account and a token that carries the matching permission.

Monitor-specific tools take a monitor_id; assistants typically get this from list_monitors first, then drill in.

Teams

If your account has more than one team, the assistant can see all of them. get_uptime_summary and get_attention_items cover every team by default, so a site that is down in a team you weren't looking at still shows up, named by team. The other read tools use your current team (the one you last opened in the dashboard) unless they are given a team: a team's slug, id or name, or all for every team. Rows in an answer that spans teams name the team they belong to.

Write tools work on one team at a time. They use the current team unless team names another one, and they never accept all. Your role on that team applies, so a viewer on a client's team can read it and cannot change it.

An API token stays on your current team unless it has the all-teams permission, which you can tick on the API settings page once you belong to two or more teams. OAuth connections act as you and reach every team.

list_teams

Every team you own or belong to, the "which team has something down?" answer.

  • Inputs: none
  • Returns: per team, id, slug, name, your role, the plan, whether it is your current team, monitor counts (total / up / down / degraded / paused), and open incidents

Read tools

get_uptime_summary

At-a-glance health across your teams, the "is anything down?" answer in one call.

  • Inputs: team, optional (defaults to every team)
  • Returns: a per-team breakdown under teams, then combined counts of monitors up / down / degraded / paused / pending (always reconcile to the total), an all_operational flag (availability only, true when nothing is down or degraded), a separate has_open_incidents flag, the lists of down and degraded monitors, and any open incidents

get_attention_items

A prioritized roundup of everything needing attention, the "anything I should know this morning?" answer.

  • Inputs: ssl_within_days (default 14), domain_within_days (default 30), regression_threshold_pct (default 50), team (defaults to every team), all optional
  • Returns: monitors down and degraded, SSL certificates and domains expiring soon, response-time regressions (recent vs prior-week average), open incidents, plus an attention_count and all_clear headline

list_monitors

List the team's monitors with their current status and recent response time.

  • Inputs: search (URL contains), status (e.g. online/offline), type (http/ping/port/heartbeat/cron), sort (response_time to rank slowest first), verbose (include full detail; default compact), limit (default 25, max 100), offset (for pagination), all optional
  • Returns: pagination fields (total, offset, limit, returned, has_more, next_offset). Each monitor is compact by default (id, url, status, 7-day avg_response_time_ms); with verbose=true it also includes name, type, check interval (seconds), and last checked

get_monitor

Full detail for one monitor, including configuration and every enabled sub-check.

  • Inputs: monitor_id (required)
  • Returns: status, per-type config (http/port/heartbeat/cron), sub_checks (SSL cert, DNS records, domain/WHOIS, keyword, JSON assertions, latest Lighthouse scores), per-region response times (or ping/port results), confirmation_seconds and recovery_seconds, blips_last_7_days (outages the confirmation period absorbed without an alert), and recent incidents

list_incidents

Incident (outage) history for the team.

  • Inputs: monitor_id, status, start_date, end_date (ISO dates), limit (default 25, max 100), offset (for pagination), all optional
  • Returns: pagination fields (total, offset, limit, returned, has_more, next_offset) plus, per incident, monitor, status, started/resolved times, duration, root cause, the diagnosis read off the failed response, and diagnostics: the response headers and traceroute Sentinel captured when the incident opened (status code, response time, every response header, and the hops to the host)

list_status_pages

The team's public status pages.

  • Inputs: search, limit, optional
  • Returns: per page, name, slug, custom domain + verification status, and the underlying monitor's status

list_groups

The team's monitor groups (the dashboard's grouped view), with ids for the group write tools and monitor group_id.

  • Inputs: none
  • Returns: per group, id, name, description, parent group, and monitor count

list_team_members

Who has access to the team, the access-review answer: "does everyone have 2FA on, and who hasn't signed in lately?".

  • Inputs: none
  • Returns: per member (owner included), name, email, role (owner/admin/editor/viewer), active or disabled status, whether two-factor auth is enabled, when they joined the team, and their last sign-in

list_expiring_certificates

SSL certificates and domain registrations expiring within a window, "any certs or domains expiring soon?".

  • Inputs: within_days (default 30, max 365), type (ssl, domain, or all), all optional
  • Returns: two lists (SSL and domains), soonest-first, url, issuer/registrar, expiry date, days until expiry, and an expired flag

get_performance_history

Trends over time for a monitor.

  • Inputs: monitor_id (required), days (default 30, max 90)
  • Returns: daily response-time trend (avg/min/max plus DNS/TCP/TLS/TTFB phases) and Lighthouse score history per strategy (performance/accessibility/SEO/best-practices + Core Web Vitals)

get_dns_history

The DNS record timeline for a monitor, reveals when records changed.

  • Inputs: monitor_id (required), record_type (e.g. A/MX/TXT), limit (default 20, max 100)
  • Returns: snapshots newest-first, record type, records, baseline flag, recorded-at

check_url

An on-demand global probe of any URL, no monitor required. Useful for "is it up everywhere?" before committing to monitoring.

  • Inputs: url (required), regions (subset of active regions), timeout_seconds (default 15, max 30)
  • Returns: an overall verdict (up / down / partial / unknown) and per-region status, HTTP code, and latency

Write tools

These change your account and require a verified email; a read-only token is refused. Your assistant will ask you to approve each call, and deletes are flagged as destructive.

create_monitor

Create a new monitor. The type must be supported by your plan.

  • Inputs: monitor_type (http/ping/port/heartbeat/cron), url, friendly_name, check_interval (minutes, 0.25 to 60, defaults to the plan floor), port, check_types, monitored_regions (defaults to all regions), and heartbeat/cron fields
  • Request options: http_method, request_headers, request_body, accepted_status_codes, follow_redirects, request_timeout, slow_response_threshold, and HTTP auth via auth_type (basic or bearer) with auth_username and auth_password (the token, for bearer; stored encrypted)
  • Alert timing: confirmation_seconds (how long the outage must hold across regions before the incident opens and alerts go out, 0 to 300, default 30, or 0 on a single-region monitor) and recovery_seconds (how long the target must stay up before the incident closes and the recovery alert goes out, 0 to 1800, default 60). 0 means at once. Not used by heartbeat and cron monitors
  • Sub-check settings: keyword_settings (up to 10 must_contain / must_not_contain rules over html or text), json_assertion_settings (up to 10 path/operator/value assertions), payment_settings (agent-payment check: HTTP 402 with a valid x402/MPP envelope is the healthy state, with optional expected amount / pay-to / network), ssl_expiry_threshold, domain_expiry_threshold

update_monitor

Update an existing monitor (type is fixed; only fields you pass change).

  • Inputs: monitor_id (required), plus any of the create fields

delete_monitor

Delete a monitor. Checks stop at once. Its history is removed for good after 7 days. Until then it can be restored from Recently deleted in the dashboard, not through MCP.

  • Inputs: monitor_id (required)

pause_monitor / unpause_monitor

Stop or resume checks and alerts for a monitor.

  • Inputs: monitor_id (required)

trigger_check

Run an on-demand check of an existing monitor now. Returns an immediate result from one location and runs a check in every region the monitor uses. Those results count like scheduled checks, so a healthy one starts the recovery period of an open incident.

  • Inputs: monitor_id (required)

create_incident

Log a manual incident against a monitor.

  • Inputs: monitor_id (required), started_at (required), status (open/resolved, required), root_cause

update_incident

Change an incident's status and/or root cause; resolving sets the end time and duration.

  • Inputs: incident_id (required), status (open/resolved, required), root_cause

resolve_incident / acknowledge_incident

Mark an incident resolved (sets end time + duration), or log an acknowledgment.

  • Inputs: incident_id (required)

delete_incident

Permanently delete an incident. Irreversible.

  • Inputs: incident_id (required)

create_group / update_group / delete_group

Manage monitor groups: the same two-level tree the dashboard shows. Deleting a group ungroups its monitors and promotes subgroups; it never deletes monitors.

  • Inputs: name (unique in team), optional description and parent_id (one level of nesting); update and delete take group_id from list_groups
  • Returns: the group's id, name, and parent

Pro Tip

These compose well: ask something like "audit example.com end to end" and the assistant will chain list_monitors, get_monitor, and get_performance_history to give a complete picture.