Manager
A Kubernetes-native control plane to install the stack, reconcile MCPServer resources, and keep deployments healthy.
Spec 00 · MCP infrastructure for Kubernetes · Alpha
Deploy, govern, and broker MCP servers using a Kubernetes-native control plane.
MCP Runtime deploys Model Context Protocol servers into your Kubernetes cluster, checks every tool call against access policy, and records each decision for audit. Agents, IDEs, and people reach your internal tools through one governed gateway.
A public preview runs at platform.mcpruntime.org. The same platform installs into your own cluster.
What it is
A control plane for the MCP servers your company runs.
MCP Runtime is a self-hosted Kubernetes control plane for internal MCP servers. It deploys and routes those servers, authorizes tool calls, expires or revokes access, and audits requests inside your cluster, next to the data and services the servers already use.
You decide who can call a tool such as delete_invoice, up to what
trust level, and for how long. Servers, grants, and sessions are Kubernetes
resources, and a gateway in each server pod enforces them on every call.
01 / Definition
One manifest becomes a governed MCP endpoint.
apiVersion: mcpruntime.org/v1alpha1
kind: MCPServer
metadata:
name: payments
spec:
image: registry.example.com/payments-mcp:v1.0.0
port: 8088
publicPathPrefix: payments
gateway:
enabled: true
analytics:
enabled: trueDeploy an MCP server the way you deploy any other service.
- Define the server, image, port, route, gateway, and analytics in one resource.
- Let the operator reconcile services, ingress, rollouts, and updates.
- Configure policy, audit, and analytics on the same resource.
02 / Architecture
Three components.
Registry
Image and deployment workflows for publishing MCP servers into environments your platform team controls.
Broker
A governed gateway that brokers agent access, applies policy before tool calls run, and records what happened.
03 / Capabilities
What it does.
Deploy servers, apply access policy, route requests through the gateway, and inspect audit and traffic data from one place.
Managed deployments
Define MCP servers once as Kubernetes CRDs. The operator handles services, rollouts, routes, and updates in a GitOps-friendly workflow.
Company access model
Grants and sessions make consent, trust level, expiry, and revocation Kubernetes resources. A call is denied unless a grant and a session allow it.
Hosted platform shape
platform.mcpruntime.org shows the platform UI as it looks after installation. Anyone can browse the preview catalog; signed-in users can publish servers to it.
Brokered tool calls
Agent traffic goes through a gateway that checks policy before a tool runs.
Operational visibility
See each MCP call with its policy decision, timing, tool name, and status in dashboards and the analytics API.
Audit-ready history
Every governed request is recorded for security review, incident response, and compliance workflows.
Platform stack included
The control plane runs in mcp-platform. Audit and telemetry run in mcp-observability. MCP servers stay in their own namespaces. Registry, ingress, API, UI, Grafana, and Prometheus install with setup.
04 / Runbook
Install the stack, then connect servers.
-
Build
go build -o bin/mcp-runtime ./cmd/mcp-runtime -
Bootstrap
./bin/mcp-runtime bootstrap -
Install
./bin/mcp-runtime setup --test-mode -
Verify
./bin/mcp-runtime status
05 / Use cases
Where companies put it to work.
Agent tool access
Expose internal tools to agents with controlled access, request inspection, and audit trails.
Internal MCP operations
Give teams one place to publish, operate, govern, and inspect MCP servers.
Try, then self-host
Try the public preview, then install the same platform in your own cluster for your own servers.
Cross-team governance
Grant another team time-limited access with a trust ceiling, and revoke it with one field.
Compliance review
Give security and compliance teams a searchable history of every tool call.
06 / Targets
Runs where your company already runs Kubernetes.
Kubernetes 1.25 or later, on any distribution. Open source and self-hosted.
k3s
kind
minikube
kubeadm
EKS
GKE
AKS
OpenShift
07 / Hosting
We can run it for you.
If you would rather not operate MCP Runtime yourself, we can host it for you on our infrastructure, or set it up and manage it in yours. Either way you get the same guardrails: deny-by-default tool policy, time-limited and revocable agent sessions, and an audit record of every tool call.
Email us with your use case and where you want it to run.
On our infrastructure
A dedicated MCP Runtime install that we run and maintain. You deploy servers and manage access; we handle the cluster, upgrades, and monitoring.
On your infrastructure
We install and manage MCP Runtime in your own Kubernetes cluster, so your MCP servers and their data stay on infrastructure you control.
Priced at cost
We charge only what it costs to run and manage the platform, with no profit margin. We are looking for our first customers: the first few pay the infrastructure cost only.
Preview the platform, then run it yourself.
Open the public platform to see the UI, or follow the install guide to run MCP Runtime in your own cluster.