MCP Servers

OpenChoreo provides Model Context Protocol (MCP) servers that enable AI assistants to interact with your OpenChoreo platform.

Overview

OpenChoreo provides two MCP servers:

1. Control Plane MCP Server - Provides tools for managing OpenChoreo resources (namespaces, projects, components, builds, deployments, infrastructure)
2. Observability Plane MCP Server - Provides tools for accessing observability data (logs, metrics, traces, alerts, incidents, and FinOps cost insights)

Each MCP server is independently accessible and requires separate configuration.

Using Both MCP Servers Together

Recommended Configuration

For the best experience, configure both the Control Plane and Observability Plane MCP servers together. Observability tools accept resource names (namespace, project, component, environment) as parameters — use the Control Plane MCP server to list and discover these names, then pass them directly to Observability tools.

Example workflow:

1. Use Control Plane MCP (list_namespaces, list_projects, list_components, list_environments) to discover resource names
2. Use those names with Observability Plane MCP tools (query_component_logs, query_traces, query_http_metrics, query_alerts) to fetch observability data

Available Tools

Control Plane MCP Server

The Control Plane MCP server is organized into seven toolsets.

Tools in multiple toolsets

Some read-only tools (marked with †) appear in both the default developer toolsets and the PE toolset. This allows platform engineers to use the PE toolset alongside Namespace and Project toolsets without needing to enable all developer toolsets.

Namespace Toolset

• list_namespaces — List all namespaces (top-level containers for organizing projects, components, and resources)
• create_namespace — Create a new namespace
• list_secret_references ‡ — List all secret references (credentials and sensitive configuration) for a namespace
• get_secret_reference ‡ — Get a single secret reference's full spec (template, data sources, refresh interval, target plane). For sync status, query get_resource_events against the rendered ExternalSecret on the data plane

‡ Also registered on the PE toolset. Authoring secret references (create_, update_, delete_secret_reference) is PE-only — see the PE toolset below.

Project Toolset

• list_projects — List all projects in a namespace (logical groupings of related components sharing deployment pipelines)
• create_project — Create a new project in a namespace
• update_project — Partially update a project (display_name, description, deployment_pipeline); only provided fields change
• delete_project — Delete a project; cascades via finalizer to remove all components, workloads, releases, and bindings owned by the project

Component Toolset

Component Management

• create_component — Create a new component in a project
• list_components — List all components in a project (deployable units with independent build and deployment lifecycles)
• get_component — Get detailed component info including configuration, deployment status, and builds
• patch_component — Partially update a component's configuration (display_name, description, autoDeploy, parameters, traits, workflow); pass an empty array for traits to clear all traits
• delete_component — Delete a component; cascades via finalizer to its workloads, releases, and release bindings
• get_component_schema — Get the JSON schema for a component's configuration options

Workload Management

• list_workloads — List workloads for a component (names, images, endpoint names)
• get_workload — Get detailed workload info; returns the full Spec (round-trip-safe with update_workload)
• create_workload — Create a new workload for a component (runtime spec: containers, resources, env vars)
• update_workload — Update an existing workload's specification (container config, env vars, port mappings, resource limits)
• delete_workload — Delete a workload; for reversible removal of just the deployment, use update_release_binding with release_state: Undeploy
• get_workload_schema — Get the JSON schema for the workload specification

Platform Standards — Read-Only

These tools are namespace-scoped by default; pass scope: "cluster" to operate on the platform-wide cluster-scoped resource.

• list_component_types † — List available component types in a namespace (WebApplication, Service, ScheduledTask, etc.)
• get_component_type † — Get the full definition of a component type including its complete spec
• get_component_type_schema † — Get the parameter schema for a component type
• list_traits † — List available traits in a namespace (autoscaling, ingress, service mesh, etc.)
• get_trait † — Get the full definition of a trait including its complete spec
• get_trait_schema † — Get the parameter schema for a trait

Resource Toolset

The Resource toolset mirrors the Component toolset for managed infrastructure: developers create Resources that reference platform-engineer-published ResourceTypes, and workloads consume the declared outputs via dependencies.resources[]. ResourceType writes, ResourceRelease management, and ResourceReleaseBinding CRUD live in the PE and Deployment toolsets respectively.

Resource Management

• list_resources — List all Resources in a project (managed-infrastructure dependencies like databases, queues, caches, object stores)
• get_resource — Get a Resource's full spec including parameters, referenced (Cluster)ResourceType, latest cut release, and conditions
• create_resource — Create a new Resource referencing a ResourceType or ClusterResourceType; the controller cuts the first ResourceRelease automatically
• update_resource — Update a Resource's parameters (full replacement). spec.owner and spec.type are immutable
• delete_resource — Delete a Resource; the two-phase finalizer blocks until all ResourceReleaseBindings are removed, then GCs owned ResourceReleases

Platform Standards — Read-Only, Namespace-Scoped

• list_resource_types † — List available ResourceTypes in a namespace (Postgres, Valkey, NATS, S3 buckets, Crossplane claims, etc.). Pass scope: "cluster" for ClusterResourceTypes
• get_resource_type † — Get the full definition of a ResourceType including parameter schema, environmentConfigs schema, outputs, and rendered resources. Pass scope: "cluster" for a ClusterResourceType
• get_resource_type_schema † — Get the parameter schema for a ResourceType. Pass scope: "cluster" for a ClusterResourceType

Deployment Toolset

Components

• create_release_binding — Create a new release binding to deploy a component to a specific environment
• list_release_bindings — List release bindings associating releases with environments for a component
• get_release_binding — Get detailed info for a specific release binding including state, overrides, and endpoints
• update_release_binding — Update a release binding's configuration (release, release_state, overrides, traits, workload settings); set release_state: Active to deploy or Undeploy to remove from the data plane while keeping the binding
• delete_release_binding — Delete a release binding entirely; for reversible removal that keeps the binding, use update_release_binding with release_state: Undeploy instead
• delete_component_release — Delete a component release (immutable; useful for pruning old releases)

Resources

• create_resource_release_binding — Create a new ResourceReleaseBinding to deploy a Resource to a specific environment
• list_resource_release_bindings — List ResourceReleaseBindings in a namespace; optionally filter by project, resource, or environment
• get_resource_release_binding — Get detailed info for a ResourceReleaseBinding including the pinned ResourceRelease, retention policy, environment overrides, and resolved outputs
• update_resource_release_binding — Update a ResourceReleaseBinding's configuration (resource_release, retain_policy, resource_type_environment_configs); promotion is performed by setting resource_release to a new ResourceRelease name
• delete_resource_release_binding — Delete a ResourceReleaseBinding; if effective retainPolicy is Retain, the finalizer holds and the data-plane state persists until the policy is flipped to Delete
• delete_resource_release — Delete a ResourceRelease snapshot (immutable; useful for pruning old releases)

Pipelines & Environments

• list_deployment_pipelines — List all deployment pipelines defining environment promotion order in a namespace
• get_deployment_pipeline — Get detailed deployment pipeline info including stages, promotion order, and environments
• list_environments † — List all environments (deployment targets like dev, staging, production) in a namespace

Build Toolset

• trigger_workflow_run — Trigger a workflow run for a component using the component's configured workflow and parameters
• create_workflow_run — Create a new workflow run by specifying a workflow name and optional parameters
• list_workflow_runs — List workflow run executions in a namespace; optionally filter by project and component
• get_workflow_run — Get detailed info for a workflow run including status, tasks, and timestamps
• get_workflow_run_status — Get the live status of a workflow run (phase, task progress)
• get_workflow_run_logs — Stream live logs for a workflow run (live-only — does not return archived logs for completed runs); optional task filter
• get_workflow_run_events — Get Kubernetes events for a workflow run (scheduling, pod-startup failures); optional task filter
• list_workflows † — List all workflow templates in a namespace (CI/CD pipelines executed on the workflow plane). Pass scope: "cluster" for ClusterWorkflows
• get_workflow † — Get the full definition of a workflow including its complete spec. Pass scope: "cluster" for a ClusterWorkflow
• get_workflow_schema † — Get the parameter schema for a workflow template. Pass scope: "cluster" for a ClusterWorkflow

Platform Engineering (PE) Toolset

note

The PE toolset is enabled by default. These tools are intended for platform administrators who manage infrastructure, environments, platform standards, and deployment pipelines. If you need to disable the PE toolset, see Configuring MCP Toolsets.

Environment Management

• list_environments † — List all environments in a namespace
• create_environment — Create a new environment in a namespace
• update_environment — Update an existing environment (display name, description, production flag)
• delete_environment — Delete an environment from a namespace

Deployment Pipeline Management

• create_deployment_pipeline — Create a new deployment pipeline defining environment promotion order
• update_deployment_pipeline — Update promotion paths and approval requirements
• delete_deployment_pipeline — Delete a deployment pipeline from a namespace

Component Releases

• list_component_releases — List all releases (immutable snapshots at a specific build) for a component
• create_component_release — Create a new release from the latest build of a component
• get_component_release — Get detailed release info including build information, image tags, and deployment status
• get_component_release_schema — Get the JSON schema for a component release's configuration options

Resource Releases

• list_resource_releases — List all ResourceReleases (immutable snapshots of Resource.spec + (Cluster)ResourceType.spec) for a Resource
• create_resource_release — Manually cut a ResourceRelease from the current Resource and ResourceType state. Typically not needed in steady state—the Resource controller cuts releases automatically on hash change
• get_resource_release — Get the full snapshot including the frozen (Cluster)ResourceType spec and the Resource parameters captured at release time

Infrastructure — Planes

These tools are namespace-scoped by default; pass scope: "cluster" to operate on the cluster-scoped plane.

• list_dataplanes — List all data planes (clusters where workloads execute) in a namespace
• get_dataplane — Get detailed data plane info including cluster details, capacity, and health
• list_workflowplanes — List all workflow planes (dedicated build infrastructure) in a namespace
• get_workflowplane — Get detailed workflow plane info including cluster details and health status
• list_observability_planes — List all observability planes providing monitoring, logging, and tracing in a namespace
• get_observability_plane — Get detailed observability plane info including observer URL and health status

Platform Standards — Read

These tools are namespace-scoped by default; pass scope: "cluster" to operate on the platform-wide cluster-scoped resource.

• list_component_types † — List available component types in a namespace
• get_component_type — Get the full definition of a component type including its complete spec
• get_component_type_schema † — Get the parameter schema for a component type
• list_traits † — List available traits in a namespace
• get_trait — Get the full definition of a trait including its complete spec
• get_trait_schema † — Get the parameter schema for a trait
• list_resource_types † — List available ResourceTypes in a namespace. Pass scope: "cluster" for ClusterResourceTypes
• get_resource_type † — Get the full definition of a ResourceType including outputs and resource manifests. Pass scope: "cluster" for a ClusterResourceType
• get_resource_type_schema † — Get the parameter schema for a ResourceType. Pass scope: "cluster" for a ClusterResourceType
• list_workflows † — List all workflow templates in a namespace
• get_workflow — Get the full definition of a workflow including its complete spec
• get_workflow_schema † — Get the parameter schema for a workflow template

Platform Standards — Creation Schemas

• get_component_type_creation_schema — Get the spec schema for creating a component type. Pass scope: "cluster" for a ClusterComponentType
• get_trait_creation_schema — Get the spec schema for creating a trait. Pass scope: "cluster" for a ClusterTrait
• get_resource_type_creation_schema — Get the spec schema for creating a ResourceType. Pass scope: "cluster" for a ClusterResourceType
• get_workflow_creation_schema — Get the spec schema for creating a workflow. Pass scope: "cluster" for a ClusterWorkflow

Platform Standards — Write

These tools are namespace-scoped by default; pass scope: "cluster" to operate on the platform-wide cluster-scoped resource.

• create_component_type — Create a new component type in a namespace
• update_component_type — Update an existing component type (full replacement)
• delete_component_type — Delete a component type from a namespace
• create_trait — Create a new trait in a namespace
• update_trait — Update an existing trait (full replacement)
• delete_trait — Delete a trait from a namespace
• create_resource_type — Create a new ResourceType. Pass scope: "cluster" to create a ClusterResourceType
• update_resource_type — Update an existing ResourceType (full replacement). Pass scope: "cluster" for a ClusterResourceType
• delete_resource_type — Delete a ResourceType. Pass scope: "cluster" to delete a ClusterResourceType
• create_workflow — Create a new workflow in a namespace
• update_workflow — Update an existing workflow (full replacement)
• delete_workflow — Delete a workflow from a namespace

Authorization — Roles

• list_authz_roles — List authz roles in a namespace
• get_authz_role — Get the full definition of an authz role
• get_authz_role_creation_schema — Get the spec schema for creating an authz role
• create_authz_role — Create a new authz role
• update_authz_role — Update an existing authz role (full replacement)
• delete_authz_role — Delete an authz role

Authorization — Role Bindings

• list_authz_role_bindings — List authz role bindings in a namespace
• get_authz_role_binding — Get the full definition of an authz role binding
• get_authz_role_binding_creation_schema — Get the spec schema for creating an authz role binding
• create_authz_role_binding — Create a new authz role binding
• update_authz_role_binding — Update an existing authz role binding (full replacement)
• delete_authz_role_binding — Delete an authz role binding

Secret References

• list_secret_references ‡ — List all secret references (credentials and sensitive configuration) for a namespace
• get_secret_reference ‡ — Get a single secret reference's full spec (template, data sources, refresh interval, target plane). For sync status, query get_resource_events against the rendered ExternalSecret on the data plane
• create_secret_reference — Create a new secret reference; spec must include template (Kubernetes Secret type) and data[] (mapping of secret keys to external store references)
• update_secret_reference — Update an existing secret reference; annotations are merged, spec is replaced wholesale when provided
• delete_secret_reference — Delete a secret reference (the underlying Kubernetes Secret is removed by the controller)

‡ Also registered on the Namespace toolset so developers can list and inspect secret references without enabling PE.

Diagnostics

• get_resource_tree — Get the rendered resource tree for a release binding
• get_resource_events — Get Kubernetes events for a specific resource in a deployment (scheduling, startup issues)
• get_resource_logs — Get container logs from a specific pod in a deployment (application errors, runtime issues)
• evaluate_authz — Evaluate one or more authz requests and return allow/deny plus the matching binding chain on a deny
• list_authz_actions — List the catalog of authz action names and the lowest applicable scope for each

Observability Plane MCP Server

The Observability Plane MCP server provides tools covering logs, traces, metrics, alerts, incidents, audit logs and FinOps cost insights:

• query_component_logs — Query runtime application logs for components (services, APIs, workers, scheduled tasks); filter by project, component, environment, time range, log levels, and search phrases
• query_workflow_logs — Query CI/CD workflow run logs capturing build, test, and deployment pipeline execution details; filter by workflow run name and task name
• query_http_metrics — Query HTTP request and latency metrics; returns time-series data for request counts (total, successful, unsuccessful) and latency percentiles (p50, p90, p99)
• query_resource_metrics — Query CPU and memory resource usage metrics; returns time-series data for usage, requests, and limits — useful for capacity planning and detecting memory leaks
• query_traces — Query distributed traces; returns a list of traces with summary information including trace ID, span count, root span details, and duration
• query_trace_spans — Query all spans within a specific distributed trace by trace ID; returns span details with timing, parent span, and service information
• get_span_details — Get full details for a specific span including attributes, resource attributes, parent span ID, and timing details
• query_alerts — Query fired alerts in OpenChoreo; supports filtering by project, component, environment, and time range — useful for investigating recent alerts and their details
• query_incidents — Query incidents in OpenChoreo; supports filtering by project, component, environment, and time range — useful for tracking incident lifecycle and response status; all incidents have an accompanying alert but not the other way around
• query_costs — Query per-component infrastructure costs (CPU cost, memory cost, and resource efficiency) for a namespace within an environment and time range; covers every component by default, or scope to a project or a single component; supports a granularity parameter to split each component into time buckets, useful for cost attribution and spotting inefficient workloads
• query_recommendations — Query right-sizing recommendations comparing current CPU/memory requests and limits against recommended values derived from observed usage, with associated costs, for a namespace within an environment and time range; covers every component by default, or scope to a project or a single component. Useful for reducing over-provisioning
• query_audit_logs — Query the audit trail: who did what, to which resource, and whether it was allowed; filter by actor (ID, type, issuer, session, entitlements), resource (type, namespace, environment, project, component, name), action, category, result, surface, operation ID, request ID and a search phrase, within a window of at most 366 days; optionally returns a timeline of record counts. Requires the cluster-scoped auditlogs:view permission. See Audit Logging

Configuring MCP Toolsets

The Control Plane MCP server exposes seven toolsets: namespace, project, component, resource, deployment, build, and pe. All seven are enabled by default. You can control which toolsets are active by passing a JSON array to openchoreoApi.config.mcp.toolsets via helm upgrade.

Syntax:

helm upgrade openchoreo-control-plane oci://ghcr.io/openchoreo/helm-charts/openchoreo-control-plane \
--version 1.3.0-rc.1 \
--namespace openchoreo-control-plane \
--reuse-values \
--set-json 'openchoreoApi.config.mcp.toolsets=["<toolset1>","<toolset2>",...]'

Common Configurations

Developer toolsets:

helm upgrade openchoreo-control-plane oci://ghcr.io/openchoreo/helm-charts/openchoreo-control-plane \
--version 1.3.0-rc.1 \
--namespace openchoreo-control-plane \
--reuse-values \
--set-json 'openchoreoApi.config.mcp.toolsets=["namespace","project","component","resource","deployment","build"]'

Platform Engineer toolset:

helm upgrade openchoreo-control-plane oci://ghcr.io/openchoreo/helm-charts/openchoreo-control-plane \
--version 1.3.0-rc.1 \
--namespace openchoreo-control-plane \
--reuse-values \
--set-json 'openchoreoApi.config.mcp.toolsets=["namespace","project","pe"]'

tip

The namespace and project toolsets are useful companions for the PE toolset — they let platform engineers discover resources without enabling all developer toolsets.

Per-Session Tool Filtering

By default, the Control Plane MCP server filters tools/list and tools/call results by the authenticated user's permissions. Tools the user lacks permission for are hidden, and unauthorized calls are rejected. MCP clients can additionally narrow what a session sees or opt out of authorization-based filtering by passing optional query parameters when establishing the session.

Query Parameter	Required	Default	Description
toolsets	false	All configured toolsets	Comma-separated list of toolsets to expose via tools/list for the session (e.g. namespace,component,pe). Unknown toolset names are silently ignored.
filterByAuthz	false	true	When true, hides tools the user cannot call and rejects unauthorized invocations. Set to false to disable MCP-layer authorization filtering.

Example:

http://api.openchoreo.localhost:8080/mcp?toolsets=namespace,component,pe&filterByAuthz=false

note

These query parameters are HTTP-only and read once when the session is established. The toolsets parameter only filters tools/list results; it does not gate tools/call. The control plane API enforces authorization independently of filterByAuthz, so disabling MCP-layer filtering does not grant additional access. It just exposes tools the user cannot successfully invoke.

Authorization of scope-collapsed tools

A scope-collapsed tool (one that accepts a scope argument) requires a distinct permission per scope — one for scope: "namespace" and one for scope: "cluster". It appears in tools/list as long as the user holds the permission for at least one scope; a tools/call invocation is then authorized against the permission for the scope actually requested. So a user with only namespace-level permissions still sees the tool, but a call with scope: "cluster" is rejected unless they also hold the cluster-level permission.

Finding MCP Server URLs

MCP server URLs follow a simple pattern based on the respective service's hostname:

• Control Plane MCP Server: <control-plane-api-hostname>/mcp
• Observability Plane MCP Server: <observer-service-hostname>/mcp

Example: Local k3d Setup

If you followed the local setup guide, your MCP server URLs will be:

Single Cluster Installation:

• Control Plane MCP Server: http://api.openchoreo.localhost:8080/mcp
• Observability Plane MCP Server: http://observer.openchoreo.localhost:11080/mcp

Multi-Cluster Installation:

• Control Plane MCP Server: http://api.openchoreo.localhost:8080/mcp
• Observability Plane MCP Server: http://observer.observability.openchoreo.localhost:11087/mcp

note

The k3d local setup examples show different ports and hostnames:

• Control plane services use port 8080
• Observability plane uses port 11080 (single cluster) or 11087 (multi-cluster)
• Multi-cluster deployments use distinct subdomain patterns for namespace isolation

Default OAuth Applications

OpenChoreo ships with pre-created OAuth applications in the default ThunderID identity provider for quick-start configuration:

Application	Client ID	Grant Type	Use Case
Browser-Based (Public)	user_mcp_client	Authorization Code + PKCE	Interactive use with AI agents that support browser-based OAuth (Claude Code, Cursor)
Service (Confidential)	service_mcp_client	Client Credentials	Programmatic access, any AI agent via bearer token

These default applications eliminate the need to manually create OAuth apps for common setups.

Next Steps

To configure your AI assistant to connect to OpenChoreo MCP servers, follow the AI Configuration Guide to set up authentication and connect your AI agent.