Skip to main content
Data Infrastructure Insights

DII MCP concepts and data model

Contributors netapp-alavoie
PDFs

Understanding these concepts helps an AI assistant select the right tools and construct valid queries.

Note The DII MCP is a Preview feature and is therefore subject to change.

Tenant scope

Every DII MCP request runs within the authenticated user's DII tenant and permissions. A tool can only return data that exists in that tenant and is visible to that identity.

Acquisition units

An acquisition unit (AU) is a remotely installed agent that polls data sources and reports telemetry to DII. AU records include connection status, reporting freshness, operating system, version, and the number of assigned data sources.

Connection and reporting are different signals. For example, an AU can accept a connection while having no successful reporting history. Assess both status and lastReported.

Data sources

A data source, also called a collector, is a configured connection to an infrastructure system such as a storage array, switch, hypervisor, or cloud service.

Use the list operation for inventory and status. Retrieve a specific data source when you need configuration details such as collection packages, poll intervals, or recent activity.

Supported data source types describe the broader DII collector catalog. They are not the same as collectors configured in the current tenant.

Objects

Objects are normalized infrastructure entities such as storage systems, pools, volumes, hosts, virtual machines, Kubernetes workloads, and product-specific resources.

Object types are dynamic. Always:

  1. List object types.

  2. Select the exact type.

  3. Retrieve metadata for that type.

  4. Query only attributes and metrics declared by the metadata.

Attributes and metrics

  • Attributes describe identity, configuration, ownership, labels, or relationships.

  • Metrics are time-varying measurements such as capacity, latency, throughput, or utilization.

  • Groupable attributes can segment an aggregate query.

Metric filters are evaluated against raw metric values before requested time-window aggregation is applied.

Logs

Log types represent event streams such as ONTAP EMS, Kubernetes events, StorageGRID events, VMware events, and DII platform events.

Log querying is also metadata-first:

  1. List available log types.

  2. Retrieve metadata for the selected type.

  3. Request only valid attributes.

  4. Search events or build an event-count histogram.

Monitors and alerts

A monitor defines a condition DII evaluates. An alert is a specific monitor result for one or more related objects.

  • Use monitor tools to inspect intent, status, object type, and corrective guidance.

  • Use alert metadata before querying alerts.

  • Use an alert name such as AL-123456 or its UUID to retrieve one alert.

Alert status and severity are separate dimensions. Ask for both when prioritizing results.

Maintenance windows

A maintenance window suppresses alerts that match an expression during a defined period. An empty successful result means that no matching windows exist; it is not evidence of missing data.

Notification rules and targets

Notification rules route matching alerts. Targets can include Slack, Microsoft Teams, PagerDuty, or custom HTTP integrations.

Webhook addresses can contain secret material. For reporting, request only the target name, engine type, and status. Never reproduce the complete address.

Metadata-first query flow

Discover types
    → retrieve type metadata
        → select valid attributes and metrics
            → query a bounded time range
                → paginate, group, or retrieve time series

Filters

DII query tools support conditions and nested logical operators:

  • Comparisons: EQ, NE, GT, LT, GE, LE

  • Pattern matching: WILDCARD

  • Presence: EXISTS

  • Logic: and, or, not

Use the exact field name and compatible value type declared in metadata. For EXISTS, omit a comparison value when supported by the connected server version.

Time ranges and pagination

  • Use ISO-8601 timestamps with explicit offsets or Z.

  • Supply an IANA timezone such as America/New_York or UTC when required.

  • Keep query windows to 30 days or less unless a tool documents a smaller limit.

  • Object and alert queries generally use limit and offset.

  • Log event queries use a returned cursor for the next page.

For errors and recovery steps, see Troubleshooting.