October 2026 release is live Read More
Breadcrumbs

Introduction

Mavvrik MCP lets supported AI clients answer cost intelligence questions using data from your connected Mavvrik tenant.

It is designed for investigation workflows where users need clear answers about spend, cost changes, anomalies, and savings recommendations. Instead of starting from a dashboard or manually adjusting filters, users can ask a question in a supported AI client and receive an answer based on their Mavvrik data.

Why this matters: a dashboard is great when you know what you're looking for and want to explore visually, but it's slow when you just have a specific question: "why did EKS spend jump last week?" means finding the right dashboard, the right filter, the right date range, before you even see an answer. MCP skips that: you ask the question in plain language, in the AI client you're already using, and get an answer grounded in your Mavvrik data.

This shows up differently depending on your role:

  • FinOps managers use this for fast, specific investigation ("what drove cost up," "are there anomalies this week") without opening a dashboard first.

  • Engineers use this to check cost impact of their own services directly from tools they already use (Cursor, Claude, ChatGPT) without switching context.

  • Finance leaders use this to get quick, conversational answers for ad-hoc questions that don't warrant building a new report.

  • Partners use this to investigate cost, usage, and savings across the child tenants they manage (directly, or rolled up across their whole portfolio) without switching between each customer's dashboard.

Mavvrik MCP can support common FinOps and engineering workflows, such as reviewing weekly cloud spend, identifying the services or accounts driving a cost increase, checking whether a spike is an anomaly, and reviewing available savings recommendations.

Current MCP coverage

Area

What it helps with

Cloud Cost

Reviewing cloud spend, identifying cost drivers, and comparing spend across services, accounts, or time periods

Cost Variances

Understanding why spend increased or decreased between two periods

Recommendations

Reviewing available savings recommendations and optimization opportunities

Anomalies

Investigating unusual cost changes or unexpected spend patterns

Cloud Resources

Reviewing cloud inventory and resource-level details across connected cloud accounts

GenAI Cost

Analyzing GenAI platform cost and AI Assist cost, including usage and spend across supported AI services

Cloud Commitments

Reviewing commitment-related insights across Reserved Instances, Savings Plans, and coverage

Kubernetes

Analyzing Kubernetes cost, cloud cost, efficiency, and utilization across clusters, namespaces, nodes, and workloads, including idle cost and GPU usage

Datacenter

Analyzing VMware datacenter cost, resource inventory, and host utilization across vCenters, datacenters, clusters, and hosts

SaaS Cost

Reviewing SaaS spend and usage data across supported SaaS integrations

Tag Explorer

Reviewing tag coverage, tag compliance, and custom tag policies across cloud resources

Cost Allocation

Reviewing allocated costs across business segments, rules, and allocation structures to understand how shared or direct costs are distributed

Budgets

Reviewing budget definitions, spend against budget, and utilization across your cloud spend

Perspectives

Reviewing saved perspectives (custom filtered cost views) and what they're scoped to

Teams

Reviewing team structure and which users belong to which teams

Users

Reviewing user accounts, roles, and status across your Mavvrik tenant

Cloud Accounts

Reviewing connected cloud provider accounts (AWS, Azure, GCP, OCI) and their connection status

Usage Accounts

Reviewing the usage/sub-account inventory under your connected cloud accounts

AI Accounts

Reviewing connected GenAI provider accounts and the models available under each

SaaS Accounts

Reviewing connected SaaS integrations and their connection status

Agentic Tasks

Reviewing agentic task activity recorded for your tenant

Agentic Pricing

Reviewing available AI providers and models used for agentic cost pricing

Agentic Price Rules

Reviewing custom markup or pricing rules applied to agentic costs

Account & Integration Settings

Reviewing tenant and personal configuration settings, and the connection status of integrations

Datacenter and Kubernetes data

The tables below show what you can ask about for datacenter and Kubernetes data, how the answers can be broken down, and how fresh the numbers are. You can combine any of the breakdowns in one question, for example "Show Kubernetes cost by namespace for the last 14 days."

Datacenter

Area

What you can ask

Dimensions

Grain

Cost

How much datacenter cost came from each cluster, host, or type of resource, and how it changed over time

vCenter, datacenter, cluster, host, resource type (virtual machine, datastore, host)

Monthly, or daily for the last 30 days

Resources

How many resources you have, and which clusters or hosts hold the most

vCenter, datacenter, cluster, host, resource type

Monthly counts, plus the current inventory for rankings

Utilization

The overall utilization score and CPU, memory, and storage utilization for your hosts, including which hosts are underused

vCenter, datacenter, cluster, host

Monthly

Kubernetes

Area

What you can ask

Dimensions

Grain

Cost

What your clusters cost, how much of it is idle, and the detail behind it down to individual pods and containers

Cluster, namespace, node, node pool, workload type, pod, container

Monthly, or a custom date range for trends

Cloud cost

What your cloud provider bills for the nodes, disks, and networking underneath your clusters

Account, cluster, resource type (node, disk, network, cluster management, load balancer), location

Monthly, or a custom date range for trends

Utilization

CPU, memory, disk, and GPU utilization for each node, including the GPU model and count

Cluster, node, location

Monthly, or a custom date range

Efficiency

How well pods use what they request: overall, CPU, memory, and GPU efficiency, with pod-level and node-level detail

Cluster, namespace, node, pod

Monthly, or a custom date range

Things to know when asking about datacenter and Kubernetes data

  • Cloud cost and datacenter cost are separate. Datacenter (VMware and on-premises) cost is reported on its own, apart from AWS, Azure, GCP, and OCI. If a cloud total looks higher than the Cost dashboard, check whether datacenter cost was included, and ask for cloud providers only to leave it out. To see datacenter cost, ask for it by name.

  • Chargeable and default cost are different figures. By default, Kubernetes and datacenter costs are shown as actual cost. Chargeable cost is the amount you would bill back to a team or application. It follows Mavvrik's chargeable-cost rules, so it is often higher and the two will not match: for Kubernetes it reflects the higher of what a workload requested and what it used, and for datacenters it is based on allocated virtual CPUs. Ask for chargeable cost only when you mean showback or chargeback.

  • Utilization and efficiency are percentages. A value of 36.8 means 36.8%. Kubernetes utilization and Kubernetes efficiency are calculated differently, so their numbers differ for the same month.

Using MCP Across Multiple Tenants (Partners)

If you're a partner managing multiple customer (child) tenants, MCP supports investigating cost data both at the individual tenant level and across your full portfolio.

List your tenants

Ask MCP which child tenants you're authorized to query, for example, "What tenants do I have access to?" This returns each tenant's name and ID, which you can reference in follow-up questions.

Query a specific child tenant

Name the tenant in your question and MCP will scope the answer to that customer specifically, for example, "What's the total cost for [tenant name] this month?" or "Are there any cost anomalies for [tenant name]?"

MCP returns the same cost figures your account would see on the corresponding Mavvrik dashboard for that tenant (including any pricing adjustment already applied, such as an Agentic Pricing Rule markup or discount). It doesn't expose a different or additional view of a tenant's cost beyond what your existing dashboard access already shows.

Query your own view (aggregated across tenants)

If you ask without naming a specific tenant, MCP answers from your own view (which, for a partner account, reflects your rolled-up portfolio across all child tenants). This is useful for portfolio-wide questions, such as "What's my total cloud spend this month?" or "Which of my tenants grew the most month-over-month?"

Example partner questions

  • "What tenants do I have access to?"

  • "What's the total cost for [tenant name] last month?"

  • "Compare cost across my child tenants this month."

  • "What savings recommendations are open for [tenant name]?"

Example questions

Examples of supported questions include:

"What drove cloud spend last week?"

"Show the top 5 services by cost this month."

"Why did EKS spend increase compared with the previous week?"

"Are there any anomalies in the last 7 days?"

"What savings recommendations are open?"

The response depends on the data available in your connected Mavvrik tenant and the access associated with your Mavvrik user account.

FAQs

What is Mavvrik MCP?

Mavvrik MCP is a read-only connection that lets compatible AI clients answer cost questions using data available in a user's Mavvrik tenant. Responses are limited by that user's Mavvrik access.

What does MCP stand for?

MCP stands for Model Context Protocol. It is an open standard that lets AI applications connect to external tools and data sources.

Does Mavvrik MCP replace the dashboard?

No. Mavvrik MCP adds a conversational interaction layer. The dashboard remains available for visual analysis, monitoring, forecasting, and policy workflows, while reports, alerts, and APIs support other ways of working with Mavvrik cost intelligence.

Does an AI client get direct access to my infrastructure?

No. Mavvrik MCP answers supported questions using data already available in Mavvrik. It does not give the AI client direct access to cloud provider accounts, Kubernetes clusters, or SaaS platforms.

Which AI clients support Mavvrik MCP?

Mavvrik provides connection guides for Claude Web, Claude Desktop, Cursor, ChatGPT, and Microsoft 365 Copilot.

How does Mavvrik MCP help with a FinOps assessment?

It helps teams review spend, changes, anomalies, savings opportunities, commitments, allocation, and ownership through questions grounded in Mavvrik data. This can reduce the manual work of exporting and reconciling reports for each assessment.