Summary
Add an optional cluster parameter to the infrastructure_health MCP tool so users can scope the health check to a single cluster (or a small set of clusters), rather than always running cluster-wide.
Background
infrastructure_health currently runs a hardcoded set of PromQL queries (cluster status, node status, aggregate status, volume status, capacity thresholds, health alerts) with no label filter — it always evaluates across every cluster the TSDB is monitoring.
In multi-cluster Harvest deployments (multiple datacenters, lab vs. prod, etc.) it's a very common ask to check the health of one specific cluster or a logical grouping of clusters (e.g. all prod clusters, one lab) without seeing noise from the others.
Proposed enhancement
Add an optional argument to the tool, e.g.:
cluster — exact cluster name, orcluster_match — regex pattern matched against the cluster label
When provided, each of the internal PromQL queries should be augmented with a {cluster="X"} (or {cluster=~"pattern"}) matcher so the report only reflects the targeted cluster(s).
When omitted, behavior is unchanged (global view).
Use cases
- "How is
prod-east-1 doing right now?"
- "Are any of my lab clusters unhealthy?"
- Per-datacenter ops dashboards / chatbot conversations scoped to a single site.
Notes
Filed as part of a NAbox chatbot integration review — all the other harvest-mcp query tools (metrics_query, metrics_range_query, list_metrics with matches, list_label_values) already allow cluster scoping via PromQL label matchers in their query arguments. infrastructure_health is the only "what's wrong?" tool that can't be narrowed.
Summary
Add an optional
clusterparameter to theinfrastructure_healthMCP tool so users can scope the health check to a single cluster (or a small set of clusters), rather than always running cluster-wide.Background
infrastructure_healthcurrently runs a hardcoded set of PromQL queries (cluster status, node status, aggregate status, volume status, capacity thresholds, health alerts) with no label filter — it always evaluates across every cluster the TSDB is monitoring.In multi-cluster Harvest deployments (multiple datacenters, lab vs. prod, etc.) it's a very common ask to check the health of one specific cluster or a logical grouping of clusters (e.g. all prod clusters, one lab) without seeing noise from the others.
Proposed enhancement
Add an optional argument to the tool, e.g.:
cluster— exact cluster name, orcluster_match— regex pattern matched against theclusterlabelWhen provided, each of the internal PromQL queries should be augmented with a
{cluster="X"}(or{cluster=~"pattern"}) matcher so the report only reflects the targeted cluster(s).When omitted, behavior is unchanged (global view).
Use cases
prod-east-1doing right now?"Notes
Filed as part of a NAbox chatbot integration review — all the other harvest-mcp query tools (
metrics_query,metrics_range_query,list_metricswithmatches,list_label_values) already allow cluster scoping via PromQL label matchers in their query arguments.infrastructure_healthis the only "what's wrong?" tool that can't be narrowed.