Skip to content
Docs

Documentation navigation

Cloud & Observability

Datadog, Sentry, PagerDuty, New Relic, Splunk, Grafana, and other ops toolsets.

#Cloud & observability

Toolsets for telemetry pipelines, incident response, on-call, and infrastructure control planes. Every connector follows the standard @toolset / @toolify shape.

#Agent-ready behavior (overview)

All list / search tools in this group return compact summaries by default with a response-local _ref field (log_ref, monitor_ref, dashboard_ref, issue_ref, incident_ref, service_ref, schedule_ref, oncall_ref, user_ref, alert_ref, app_ref, index_ref, search_ref, folder_ref, component_ref, subscriber_ref, zone_ref, record_ref, worker_ref, bucket_ref, pod_ref, namespace_ref, deployment_ref). Raw provider IDs are hidden unless you pass include_ids=True. Where the tool exposes include_metadata=False, that returns the raw provider response untouched. Default page size is 25; some tools cap absolute limits to provider maxima (1000 logs, 100 users, etc.).

Write tools that take an identifier accept either a raw ID/UID/int or the dict from the corresponding list/get tool. This includes get_monitor, delete_monitor, mute_monitor, get_issue, update_issue, delete_issue, list_events_for_issue, get_incident, update_incident, acknowledge_incident, delete_alert_policy, get_dashboard_by_uid, delete_dashboard_by_uid, update_component, update_incident, delete_incident, get_zone, list_dns_records, update_dns_record, delete_dns_record, purge_cache.

Destructive tools across this group (require user confirmation): delete_monitor, delete_issue, cancel_search_job, delete_alert_policy, delete_dashboard_by_uid, delete_incident, delete_dns_record, purge_cache, delete_pod.

#DatadogToolSet

Python 
from maivn_tools import DatadogToolSet

connector = DatadogToolSet(
    api_key=secrets["DD_API_KEY"],
    app_key=secrets["DD_APP_KEY"],     # required for most read APIs
    site="datadoghq.com",
)

Tools: search_logs, submit_logs, query_metrics, submit_metrics, list_monitors, get_monitor, create_monitor, delete_monitor (destructive), mute_monitor, post_event, list_dashboards, get_dashboard.

Agent-ready behavior

  • search_logs (default limit 25, hard max 1000; summaries capped at 25): summary with log_ref, timestamp, service, status, host, message, optional tags. Raw Datadog event IDs hidden unless include_ids=True; pass include_metadata=False for the raw provider response. Cursor pagination preserved as next_cursor.
  • list_monitors (default page size 25): summary with monitor_ref, name, type, state, message, tags. Numeric IDs hidden by default; opt in with include_ids=True (needed for get_monitor, delete_monitor, mute_monitor).
  • list_dashboards: summary with dashboard_ref, title, description, layout_type, url. Opt in with include_ids=True for dashboard_id (needed for get_dashboard).
  • Tolerant inputs: get_monitor, delete_monitor, and mute_monitor accept an int, a monitor dict, or a list of such.
  • Destructive: delete_monitor. Confirm before invoking.
Python 
hot = connector.search_logs(
    query="status:error service:checkout",
    from_time="now-15m",
    to_time="now",
)

#SentryToolSet

Python 
from maivn_tools import SentryToolSet

connector = SentryToolSet(auth_token=secrets["SENTRY_TOKEN"])

Tools: list_organizations, list_projects, get_project, list_issues, get_issue, update_issue, delete_issue (destructive), list_events_for_issue, list_releases, create_release, list_alert_rules.

Agent-ready behavior

  • list_issues (default limit 25, hard max 100; summaries capped at 25): summary with issue_ref, title, culprit, status, level, count, last_seen, permalink. Raw Sentry issue IDs hidden; opt in with include_ids=True (needed for get_issue, update_issue, delete_issue). include_metadata=False returns the raw provider payload.
  • list_alert_rules: summary with alert_ref, name, environment, frequency, conditions_count, actions_count. Opt in for alert_id.
  • Tolerant inputs: get_issue, update_issue, delete_issue, and list_events_for_issue accept a string ID, an issue dict, or a list of issue dicts.
  • Destructive: delete_issue. Confirm before invoking.

#PagerDutyToolSet

Python 
from maivn_tools import PagerDutyToolSet

connector = PagerDutyToolSet(
    api_key=secrets["PD_TOKEN"],
    from_email="oncall@example.com",        # required for ack / resolve writes
    events_routing_key=secrets["PD_RK"],   # required for Events v2 trigger / resolve
)

REST + Events v2 in one toolset. Tools: list_incidents, get_incident, update_incident, acknowledge_incident, create_incident, trigger_event, resolve_event, list_services, list_schedules, list_on_calls, list_users.

Agent-ready behavior

  • list_incidents (default limit 25, hard max 200; summaries capped at 25): summary with incident_ref, incident_number, title, status, urgency, created_at, service_name. Raw incident IDs hidden; opt in with include_ids=True for incident_id (needed for get_incident, update_incident).
  • list_services (default 25): summary with service_ref, name, description, status. Opt in for service_id (needed by create_incident).
  • list_schedules (default 25): summary with schedule_ref, name, description, time_zone. Opt in for schedule_id (needed by list_on_calls(schedule_ids=...)).
  • list_on_calls: summary with oncall_ref, user_name, escalation_level, schedule_name, start, end. Opt in for user_id / schedule_id. Best for "who is on call?" questions.
  • list_users (default 25): summary with user_ref, name, email, role, time_zone. Opt in for user_id.
  • Tolerant inputs: get_incident, update_incident, and acknowledge_incident accept a string ID, an incident dict, or a list of such.
  • No destructive tools -- incident deletion is intentionally not exposed (use update_incident(status="resolved") instead).
Python 
hot = connector.list_incidents(statuses=["triggered", "acknowledged"])
connector.acknowledge_incident(hot["incidents"][0])  # dict accepted

#NewRelicToolSet

Python 
from maivn_tools import NewRelicToolSet

connector = NewRelicToolSet(
    api_key=secrets["NEW_RELIC_USER_KEY"],
    account_id=int(secrets["NEW_RELIC_ACCOUNT_ID"]),
    ingest_license_key=secrets.get("NEW_RELIC_LICENSE_KEY"),
    region="US",
)

NerdGraph + REST + Insights ingest. submit_logs / submit_metrics require ingest_license_key. Tools: nerdgraph_query, nrql_query, submit_logs, submit_metrics, list_alert_policies, create_alert_policy, delete_alert_policy (destructive), list_applications.

Agent-ready behavior

  • list_alert_policies: summary with alert_ref, name, incident_preference, created_at. Raw numeric policy IDs hidden; opt in with include_ids=True for policy_id (needed by delete_alert_policy). include_metadata=False returns the raw response.
  • list_applications: summary with app_ref, name, language, health_status, reporting. Opt in for application_id.
  • Tolerant inputs: delete_alert_policy accepts an int, a policy dict, or a list of such.
  • Destructive: delete_alert_policy. Confirm before invoking.

#SplunkToolSet

Python 
from maivn_tools import SplunkToolSet

connector = SplunkToolSet(
    token=secrets["SPLUNK_TOKEN"],
    base_url="https://splunk.example.com:8089",
)

REST API for the splunkd management port. Tools: create_search_job, get_search_job, get_search_results, cancel_search_job (destructive), list_indexes, list_saved_searches, create_saved_search, update_saved_search.

Agent-ready behavior

  • list_indexes: summary with index_ref, name, total_event_count, current_db_size_mb, disabled (capped at 25). include_metadata=False returns the raw response.
  • list_saved_searches: summary with search_ref, name, search, is_scheduled, cron_schedule (capped at 25). Saved-search names are user-facing and are kept (they are also the API key used by update_saved_search).
  • Destructive: cancel_search_job. Confirm before invoking (results are unrecoverable once cancelled).

#GrafanaToolSet

Python 
from maivn_tools import GrafanaToolSet

connector = GrafanaToolSet(
    api_key=secrets["GRAFANA_TOKEN"],   # service-account or API key
    base_url="https://stack.grafana.net",
)

Tools: search_dashboards, get_dashboard_by_uid, create_or_update_dashboard, delete_dashboard_by_uid (destructive), list_folders, create_folder, list_datasources, create_annotation, list_alert_rules, get_health.

Agent-ready behavior

  • search_dashboards (default limit 25, hard max 5000; summaries capped at 25): summary with dashboard_ref, title, type, url, folder_title, tags. Dashboard UIDs hidden; opt in with include_ids=True for dashboard_uid / dashboard_id (needed by get_dashboard_by_uid, delete_dashboard_by_uid).
  • list_folders (default limit 25): summary with folder_ref, title, parent_uid. Opt in for folder_uid / folder_id.
  • list_alert_rules: summary with alert_ref, title, folder_uid, rule_group, condition, no_data_state, exec_err_state. Opt in for alert_uid.
  • Tolerant inputs: get_dashboard_by_uid and delete_dashboard_by_uid accept a UID string or a dashboard dict from search_dashboards(include_ids=True).
  • Destructive: delete_dashboard_by_uid. Confirm before invoking.

#StatuspageToolSet

Python 
from maivn_tools import StatuspageToolSet

connector = StatuspageToolSet(
    api_key=secrets["STATUSPAGE_KEY"],
    page_id="my-page-id",
)

Tools: get_page, list_components, update_component, list_incidents, create_incident, update_incident, delete_incident (destructive), schedule_maintenance, list_subscribers.

Agent-ready behavior

  • list_components: summary with component_ref, name, status, description, group_id. Raw component IDs hidden; opt in with include_ids=True for component_id (needed by update_component).
  • list_incidents (default per_page 25, hard max 100; summaries capped at 25): summary with incident_ref, name, status, impact, created_at, resolved_at, shortlink. Opt in for incident_id (needed by update_incident, delete_incident).
  • list_subscribers (default 25): summary with subscriber_ref, mode, email, phone_number, endpoint, quarantined_at. Opt in for subscriber_id.
  • Tolerant inputs: update_component, update_incident, and delete_incident accept a string ID or a resource dict.
  • Destructive: delete_incident. Confirm before invoking.

#CloudflareToolSet

Python 
from maivn_tools import CloudflareToolSet

connector = CloudflareToolSet(api_token=secrets["CLOUDFLARE_TOKEN"])

v4 REST API. Tools: verify_token, list_zones, get_zone, list_dns_records, create_dns_record, update_dns_record, delete_dns_record (destructive), purge_cache (destructive), list_workers, list_r2_buckets, create_r2_bucket.

Agent-ready behavior

  • list_zones (default per_page 25, hard max 50): summary with zone_ref, name, status, plan. Zone IDs hidden; opt in with include_ids=True for zone_id (needed for list_dns_records, purge_cache, etc.).
  • list_dns_records (default per_page 25, hard max 100): summary with record_ref, type, name, content, ttl, proxied. Opt in for record_id (needed for update_dns_record, delete_dns_record).
  • list_workers: summary with worker_ref, name, created_on, modified_on, etag. Opt in for script_id.
  • list_r2_buckets: summary with bucket_ref, name, creation_date, location. Bucket names are the primary identifier in R2 so no include_ids flag is needed.
  • Tolerant inputs: get_zone, list_dns_records, update_dns_record, delete_dns_record, and purge_cache accept a string ID or a resource dict from the corresponding list call.
  • Destructive: delete_dns_record, purge_cache. Confirm before invoking (purge causes user-facing latency until cache repopulates).

#KubernetesToolSet

Python 
from maivn_tools import KubernetesToolSet

connector = KubernetesToolSet(
    api_server="https://my-cluster:6443",
    token=secrets["K8S_SA_TOKEN"],
)

Talks to the Kubernetes API server directly. Tools: get_version, list_namespaces, list_pods, get_pod, get_pod_logs, list_deployments, scale_deployment, delete_pod (destructive), list_services, list_events, list_nodes, apply_manifest.

Agent-ready behavior

  • list_namespaces: summary with namespace_ref, name, phase, created_at. Namespace name is the user-facing identifier and is kept; cluster UID is opt-in via include_ids=True.
  • list_pods (default limit 25, hard max 1000; summaries capped at 25): summary with pod_ref, name, namespace, phase, ready, node, start_time. Pod UIDs are opt-in.
  • list_deployments: summary with deployment_ref, name, namespace, replicas, ready_replicas, available_replicas. UIDs opt-in.
  • list_services: summary with service_ref, name, namespace, type, cluster_ip, ports. UIDs opt-in.
  • list_events returns the raw event list (already useful for diagnosing Pending/Failed pods); cap with limit (default 25, hard max 1000).
  • Destructive: delete_pod. Confirm before invoking (one-off pods are not respawned).
apply_manifest uses server-side apply (Content-Type: application/apply-patch+yaml + fieldManager) so the caller can ship arbitrary resources without crafting per-kind tooling.