Kubeadapt MCP
Tools reference
Every Kubeadapt MCP tool: what it answers, the arguments it takes, and the kind of data a connection needs for it. All 23 tools are read-only.
Kubeadapt MCP offers 23 tools. All of them only read. Your AI tool chooses which to call; you don't need to name them, though you can.
A connection sees the tools for the kinds of data it was given when you signed in. The three core tools are always available.
What every answer contains
| Field | What it holds |
|---|---|
summary | A one-line answer in plain words |
data | The figures behind it, with units in the field names, such as cost_usd_per_month or memory_used_gib |
provenance | The source view, the time window, the cost mode, and how the figures were obtained: measured, estimated, preliminary, partial or stale. lower_bound is set when part of the cost could not be priced. |
notes | Caveats to pass on, such as a cluster that stopped reporting |
open_in_kubeadapt | A link to the same view in the app |
Money is in US dollars. Dates are calendar days in UTC, and end_date is included. Unless you give dates, cost tools read the last 30 complete days.
Core
| Tool | What it answers |
|---|---|
get_context | Your organization, your role, the kinds of data the connection may read, and every cluster it can see, with provider, size, status and monthly run rate. AI tools call it first. |
search | Clusters, namespaces, workloads, nodes, teams and departments whose name contains some text, with ids the other tools take. |
fetch | One record by id: a cluster, namespace, workload, node, team, department, recommendation or incident. |
Costs and carbon
| Tool | What it answers | Main arguments |
|---|---|---|
query_costs | Spend over a period, grouped by cluster, namespace, workload, workload kind, node, pod or container, with the previous period, the change, the top rows and their run rate. cost_split divides the bill into workload usage, idle capacity and shared cost. | group_by, start_date, end_date, cluster_ids, namespaces, teams, departments, label_key, label_value, cost_mode, sort_by |
get_cost_trend | Daily, weekly or monthly spend for the organization, a cluster, a namespace or a workload, and an estimate of where the current month will end. | cluster_id, namespace, workload_uid, granularity, project_month_end |
compare_costs | Two periods side by side and the rows that moved most. | group_by, start_date, end_date, compare_start_date, compare_end_date |
get_team_costs | Spend by team and department for a month, a team's or department's detail with its projected month-end, or the workloads no team owns. | month, team_id, department_id, unassigned |
get_pricing | The rates a cluster is priced with and where each comes from (custom, the cloud's list price, or Kubeadapt's defaults): compute, storage, management fee and network per traffic class. | cluster_id |
list_volumes | Persistent volumes, most expensive first, with size, storage class, status, claim and monthly cost; unattached volumes called out. | cluster_id, namespace, status, storage_class |
list_cost_labels | The Kubernetes label keys on your workloads, or one key's values, to filter query_costs by. | key, search, cluster_ids |
get_carbon_footprint | Emissions in kg CO₂e and energy in kWh by cluster, namespace, workload and node. | cluster_id, start_date, end_date |
Savings
| Tool | What it answers | Main arguments |
|---|---|---|
list_recommendations | Rightsizing, spot, abandoned-workload and orphaned-resource recommendations with their monthly saving. | cluster_id, type, namespace, status |
get_rightsizing_plan | One workload's recommended requests and limits per container, p95 and peak usage, OOM kills and restarts, the saving, and a YAML patch to review. | cluster_id, workload_uid or namespace and workload_name |
get_capacity_overview | A cluster's CPU and memory capacity, requests and use, unrequested capacity, growth and headroom, node groups and the most expensive namespaces. | cluster_id |
list_nodes | Nodes or node groups with instance type, spot or on-demand, allocatable, requested and used CPU and memory, pods and monthly cost; least utilized first on request. | cluster_id, group_by, sort_by, capacity_type |
get_usage_history | CPU and memory used against requests and limits over time for a workload, namespace or node, with averages, p95, peaks and the busiest hour of the day. | cluster_id and one of workload_uid, namespace, node; start_date, end_date |
get_node_optimization | Kubeadapt's recommended node layout and its saving, or a what-if for moving capacity to spot, adding or removing nodes, or moving to ARM instances. | cluster_id, spot_percent, node_scaling_pct, arm_migration_pct |
list_best_practice_findings | Configuration findings such as missing requests, limits or probes, with the compliance score. | cluster_id, severity, category, namespace |
Alerts
| Tool | What it answers | Main arguments |
|---|---|---|
list_incidents | Open and recent incidents with their values and, for budgets, the share of the budget spent. | state, alert_type, since |
get_incident | One incident's timeline, findings about the cause, and cost impact. | incident_id |
list_alert_rules | Alert rules and budgets with their thresholds, state, recent incidents and budget use this period. | alert_type, state |
Network costs
| Tool | What it answers | Main arguments |
|---|---|---|
query_network_costs | Traffic and its cost by billing class, internet egress, cross-zone traffic, the workloads that send most, top destinations and, on request, the zone-to-zone matrix. | start_date, end_date, cluster_ids, include_zones |
get_workload_traffic | Who one workload talks to, in which direction, across which zones, and what that traffic costs per month. | cluster_id, workload_uid |
Network costs are available for organizations that use Network Intelligence. When a cluster in scope has no network data, the answer says its traffic is unknown rather than reporting zero.
Prompts and resources
Kubeadapt MCP also offers six prompts (see Ready-made prompts) and a glossary resource, kubeadapt://glossary, that defines run rate, fully loaded cost, idle cost, shared cost and the measurement statuses.