Open source MCP platform for Kubernetes

Deploy, govern, and observe MCP services on Kubernetes.

MCP Runtime gives platform teams a higher-level control plane for MCP delivery, access, policy, audit, and observability.

Documentation Runtime guide

Open source Kubernetes-native Governed request path

Reference manifest

One service definition. One operating model.

The runtime owns rollout and routing from this service definition. Sentinel governs the live request path with policy, audit, and observability.

Runtime

MCP platform control plane

Bootstrap, reconciliation, rollout, ingress, and lifecycle.

Access

Grants + sessions stay explicit

Consent, trust ceilings, expiry, and revocation stay first-class.

Sentinel

Governed MCP request path

Policy, audit, and observability on live MCP requests.

apiVersion: mcpruntime.org/v1alpha1
kind: MCPServer
metadata:
  name: payments
spec:
  image: registry.example.com/payments-mcp:v1.0.0
  port: 8088
  ingressHost: mcp.example.com
  ingressPath: /payments/mcp
  gateway:
    enabled: true
  analytics:
    enabled: true

Alpha

Runtime, access, and the governed request path already work end to end. The architecture is stable enough to evaluate. The API and UX are still evolving.

Start with the page that matches the task

Documentation is the fastest way to understand the platform boundary and the operator workflow.

Runtime

Control plane and service lifecycle

See cluster bootstrap, reconciliation, rollout, ingress, and delivery state.

Runtime docs

CLI

Commands operators run day to day

Start with setup, cluster, server, registry, pipeline, and status workflows.

CLI docs

Sentinel

Governed request path and observability

Review proxy enforcement, audit events, query APIs, and observability.

Sentinel docs

API

The resource and request contract

Use the API reference for YAML examples, field semantics, headers, and status.

API reference

Three jobs, one platform boundary.

Keep lifecycle, grants, and request governance distinct without splitting them across unrelated tools.

Runtime

Control plane for MCP services

Own setup, registry, ingress, reconciliation, and rollout from one Kubernetes-native surface.

Cluster bootstrap Reconciliation Rollout + routes

Sentinel

Governed request path with policy and observability

Put enforcement, audit, and telemetry on live MCP requests instead of rebuilding them inside every service.

Proxy enforcement Audit + telemetry UI + APIs

Access

Explicit grants and sessions

Keep entitlement, consent, trust, and revocation in dedicated resources instead of app-specific conventions.

Entitlement Consent + expiry Revocation

From service definition to governed traffic

Follow the path from definition to rollout, enforcement, and audit without leaving the platform model.

Define once

Describe image, route, gateway, analytics, and access expectations in one runtime definition.

Reconcile

Use the CLI and operator to prepare cluster state and expose the MCP service through a stable path.

Govern live requests

Route requests through the proxy path when gateway mode is enabled so identity, policy, audit, and telemetry happen in one place.

Inspect and iterate

Use grants, sessions, and sentinel surfaces to review behavior and tighten policy as the service evolves.

Four surfaces, one documentation model

Move from architecture to day-two operations without switching mental models.

Runtime setup / cluster / operator / MCPServer

Prepare clusters and keep MCP services reconciled.

Access MCPAccessGrant / MCPAgentSession

Keep entitlement, consent, and revocation explicit.

Sentinel proxy / gateway / ingest / processor / api / ui

Handle live request governance, audit, and observability.

Docs runtime / cli / sentinel / api

Move from architecture to exact fields and commands quickly.

Start with the docs

Start with runtime docs, then follow the request path.

Read the runtime first for lifecycle and delivery, then Sentinel for policy, audit, and observability on live MCP requests.

Sentinel docs Runtime docs