Use these tools when you want the LLM to calculate live metric aggregations from Argus APIs.
The LLM should first use the Metric Search Translator MCP tools to shape and validate the aggregation request.
ProcessingEventsPerFilter metrics cover annotation filters, analysis filters, and match filters.
At least one statistics.values item is required for a meaningful aggregation.
Metric aggregation responses are source-of-truth results. The LLM should explain returned metric values only, without inventing missing buckets, customers, timestamps, units, or key translations.
Execute an aggregation for ProcessingEventsPerFilter metric records. The result contains aggregated metric values for the selected filters, groupings, and time buckets.
Arguments
Field
Description
Type
Default
Required
request
ProcessingEventsPerFilter metric aggregation definition prepared and validated from the user's request.
Bucket size as a number using resolutionUnit. Use 0 to aggregate the full requested time range into one bucket. Values above 10000 are capped at 10000.
number
0
No
resolutionUnit
Unit for resolution. Allowed values: milliseconds, seconds, minutes, hours, days.
Each item in data is one aggregate result, optionally grouped by customer, descriptor keys, and time bucket.
Important response fields:
customer: customer associated with the aggregate, if present.
keys: grouping key values for the aggregate. Each key can contain name, raw value, and optional translation.
metricTimestamp: timestamp for the aggregate bucket when time bucketing was requested.
values: aggregated metric values, with metric name, numeric value, and optional unit.
metricDescriptorID: descriptor identifier. It is not a user-facing metric value.
Usage Notes
The LLM should use generateProcessingEventsPerFilterMetricAggregationQuery before executing the aggregation so metric values, grouping, time bucketing, and enum values are validated.
Use sum for totals, avg for arithmetic averages, min and max for extremes, and wavg only when the user asks for or needs a weighted average.
When wavg is used, weightByValue must identify the metric value that supplies the weight.
For time bucketing, choose a resolution that produces at most 10000 buckets over the selected time range.
Use translation for display when available, but preserve raw key values when exact identifiers matter.
Do not combine or infer aggregate records that are not present in the response.