Real‑time cost allocation
Break down costs by Kubernetes concepts (CPU, GPU, memory) to the container level. Find waste in minutes, by namespace, workload or node.
Coming soon — MCP server for Claude, Gemini & GPT
Measure Kubernetes cost with eBPF. Ask about it with AI.
KubeSpend is a vendor‑neutral, eBPF-powered platform for measuring and allocating cloud infrastructure and container costs in real time — down to the cross-zone and egress traffic your billing console cannot see. Our MCP server lands soon, so the model you already work in can query all of it directly.
Used by SREs and FinOps teams to drive immediate savings
- Network visibility
- Cross-zone & egress
- Lightweight agent
- Flat pricing
- MCP server — coming soon
console.kubespend.io
Features
Cost data that lines up with how you run Kubernetes.
Allocation down to the container, network cost the billing console cannot show you, and recommendations you can act on the same day — in the console, over the API, or soon through your AI assistant.
Cross-zone & egress costs
eBPF flow metadata shows exactly which services drive inter-zone and egress traffic, priced from live rates or your own rate card.
In-cluster & out-of-cluster
Track provider resources like object storage, load balancers and databases alongside cluster costs, in one total.
Enterprise-ready & secure
Maintained by experts. API-key ingest, OTP or Google sign-in, and a self-hosted install when data cannot leave your perimeter.
AI-native, through MCP Coming soon
eBPF collects the data no billing export has; the MCP server hands it to the model you already work in. Claude, Gemini, GPT, Cursor or Kiro can read your cost, efficiency and network figures and answer in a sentence.
eBPF + AI · MCP server
Coming soonAsk your cluster what it costs — from the model you already use.
eBPF is how we get cost data nobody else has: per-service cross-zone and egress bytes, priced. The MCP server is how you use it without a dashboard. Point Claude, Gemini, GPT, or any Model Context Protocol client at your workspace and cost, efficiency, network flows and rightsizing become something you ask about in a sentence.
Tools the model gets
Read-only. Scoped to one API key, one organization.
- Cluster and namespace cost over any date range, split by compute, network and storage.
- Workload efficiency — requests versus actual usage, so idle capacity has a number attached.
- Network flows from the eBPF agent, with cross-zone and egress bytes priced per service edge.
- Savings recommendations and budget burn rate, ranked the way the console ranks them.
No write tools at launch. The MCP server answers questions about your spend; it does not resize workloads, edit budgets or touch your cluster.
How you will connect
Setup instructions publish in the docs when it ships.
- 1 Add KubeSpend as one server entry in your MCP client's configuration.
- 2 Give it an API key from the console. It sees only that key's organization, and only to read.
- 3 Self-hosted? Point it at your own API, so the model talks to your instance, not ours.
Works with any MCP client:
Claude
Gemini
GPT
Cursor
Kiro
What that looks like in practice
“Which namespace grew the most this month, and what changed?”
“Show me the five service pairs driving cross-zone traffic and what they cost.”
“Draft the rightsizing PR description for the top three overprovisioned deployments.”
Get notified when it ships
What is MCP?
Mention “MCP” in the form and we will add you to the preview list.
In the console
Six screens, one question: where is the money going?
Every panel below is a page in the KubeSpend console. Connect a cluster and they fill in within a minute.
Dashboard
Spend, waste and agent health across every connected cluster, with month-over-month movement.
Cost Analysis
Stacked cost over any date range, split by compute, network and storage. Export to CSV for finance.
Namespaces & workloads
Per-namespace and per-workload cost with request-vs-usage efficiency, so idle capacity is obvious.
Network traffic
A service-to-service map from eBPF flow data, with cross-zone and egress bytes priced per edge. Requires the eBPF add-on.
Available savings
Rightsizing recommendations with the monthly figure attached, ranked so you start with the largest one.
Budgets
Per-cluster and per-namespace budgets with burn-rate tracking, so an overspend is caught mid-month.
Already have an account? Sign in to the console →
How it works
Install the agent in ~60 seconds.
One Helm release, one API key. No cloud credentials, no billing account access, no CRDs to reconcile.
1
Create an API key
Sign in to the console and create a key for your organization. Self-hosted deployments use their own console URL.
Settings → API keys → Create key
2
Install via Helm
The cluster name is auto-detected, so the key is the only value you have to provide. The chart is pulled from Amazon ECR Public; Helm 3.8+.
kubectl create namespace kubespend
kubectl -n kubespend create secret generic kubespend-creds \
--from-literal=api-key="<YOUR_API_KEY>"
helm install kubespend-agent \
oci://public.ecr.aws/kubespend.io/charts/kubespend-agent \
--namespace kubespend --set existingSecret=kubespend-creds
3
Instant visibility
Your cluster appears in the console within a minute, with cost and efficiency populated from the first scrape. Then ship rightsizing recommendations, not guesses.
- Kubernetes 1.24+ with metrics-server
- Helm 3.8+
- Linux nodes on kernel 6.6+ for the eBPF add-on
Adding network cost visibility
The eBPF DaemonSet attaches TC programs to pod interfaces and reports flow metadata —
never payloads. On kernel 6.6+ it runs with BPF, PERFMON and NET_ADMIN capabilities
rather than privileged: true.
Self-hosted or SaaS — your security, your choice.
Choose the deployment model that fits your compliance needs. The same console, the same agents, the same data model either way.
Runs inside your VPC
Console, API and storage stay within your perimeter.
Token-based auth
Scoped API keys for ingest, OTP or Google SSO for people. No shared credentials.
No cloud-billing ingestion
Zero access to your billing accounts. Pricing comes from public rates or your own rate card.
Data residency controls
Pick the region, or keep everything on infrastructure you own.
Pricing
Flat, per cluster. Never a cut of your savings.
Pricing scales with cluster size, so the bill is predictable before you install anything. Optimising harder never costs you more.
Core
Cost allocation and rightsizing for every connected cluster.
- Cost by namespace, workload and node
- Rightsizing recommendations
- Budgets and CSV export
- Lightweight Deployment agent
Most
popular
Core + Network
Everything in Core, plus the eBPF network cost add-on.
- Cross-zone and egress cost per service
- Service-to-service traffic map
- Zone-aware routing opportunities
- eBPF DaemonSet, metadata only
Self-hosted
The full stack inside your own VPC, for teams with residency requirements.
- Console, API and storage you operate
- No data leaves your perimeter
- Custom rate cards
- Priority support from the maintainers
Early access pricing is set per cluster size — ask us for a quote.
MCP server access is included on every tier when it ships. There is no AI add-on and no per-query charge — it reads the data your plan already collects.
Why KubeSpend?
Unlike other cloud optimisation tools, which focus on compute, KubeSpend optimizes the hidden 40% of your bill: network. Simple flat pricing, and it runs securely in your VPC.
Where other tools require heavy integration or cloud billing access, KubeSpend delivers value from metadata alone. Find idle requests. Reclaim compute. Reduce bills.
And it is built to be read by more than a browser. eBPF gives us cost data no billing export contains; the MCP server — coming soon — puts that data inside the assistant your team already asks questions in, so the analysis happens where the work happens.
Frequently asked questions
How does KubeSpend calculate network cost?
We ingest network flow metadata to identify cross-zone and egress traffic, applying standard cloud pricing or custom rate cards.
Can I self-host?
Yes. You can self-host KubeSpend inside your VPC for maximum security and data residency control.
Is there agent overhead?
The agent is extremely lightweight, consuming minimal resources to ingest metadata without impacting workload performance.
Does it use cloud billing?
No cloud-billing ingestion is required. We use standard cloud pricing or custom rate cards.
What is the pricing model?
We use a simple flat pricing model based on your cluster size. We do not charge a percentage of your savings.
Can I connect Claude, Gemini or GPT to my cost data?
That is what the KubeSpend MCP server does, and it is coming soon. Any Model Context Protocol client — Claude, Gemini, GPT, Cursor, Kiro — reads the same cost, efficiency and eBPF network data the console renders. The tools are read-only and scoped to one API key, so the model can answer questions about your spend but cannot change anything, and self-hosted deployments point the client at their own API.
How do I get started?
Install the agent via Helm in ~60 seconds and connect with a token-based login to see results instantly.
Start saving today
Ready to cut your Kubernetes costs? Request early access and we will get you a key.
No credit card · 1-minute token login · self-host or hosted
Early access or pricing
Tell us about the clusters you want to measure, or ask for a quote. We reply within one business day.
Already a customer?
Support cases are filed from the console after you sign in, so they reach us with your organization and clusters already attached.