Loading ToolBench

Reading the current corpus…

k8s-mcp MCP server · F · ToolBench

← Browse

k8s-mcp

github.com/AceDarkknight/k8s-mcp

Kubernetes MCP Server - provides Model Context Protocol access to Kubernetes cluster management and resource operations

2026-07-28+STREAMABLE_HTTP13 tools

47/100

Definition Quality60

Protocol Readiness55

Supportability21

Spec posture

Inferred effective spec 2026-07-28+·current baseline 2026-07-28

Static source inference · medium confidence · detected: Logging

Deprecated protocol patterns detected

Summary

This Kubernetes MCP server provides 13 READ_ONLY tools with consistent naming (all verb_noun: get_, list_, check_). However, tool definitions suffer from significant schema incompleteness and minimal parameter documentation. While tool names follow verb-starting conventions and descriptions exist, the schema definitions visible in the source code lack proper JSON Schema structures with type declarations and constraint definitions. Input schemas are provided in the evaluation metadata but are NOT visible in the actual source code excerpts shown (internal/mcp/server.go shows tool registration without input schemas). No output schemas are documented anywhere. Error handling exists (per features list) but guidance on recovery or categorization is not evident in the visible code. The server uses HTTP transport (current standard, good), but lacks modern MCP features like tool annotations, progress reporting, and structured error classification.

Tools (13)

check_rbac_permissionread onlyauthsource verified62/100

Check if the current user has permission to perform an action (kubectl auth can-i). Parameters: verb (string, required, e.g. 'get', 'list'), resource (string, required, e.g. 'pods'), namespace (string, required)

get_cluster_statusread onlyauthsource verified58/100

Get cluster status information (version, node count, namespace count)

get_eventsread onlyauthsource verified58/100

Get cluster events. Parameters: namespace (string, required)

get_pod_logsread onlyauthsource verified65/100

Get pod logs. Default tail_lines=100, max_bytes=1MB. Parameters: pod_name (string, required), namespace (string, required), container_name (string, optional), tail_lines (int, optional), previous (bool, optional), cluster_name (string, optional)

get_resourceread onlyauthsource verified58/100

Get detailed information about a specific resource (JSON format). Secrets will be redacted. Parameters: resource_type (string, required, e.g. 'pods' or 'pod'), name (string, required), namespace (string, required)

get_resource_yamlread onlyauth

Issues (7)

CRITICALpattern:constrained-input ↗

Input schemas not visible in source code. Tool registration in internal/mcp/server.go uses mcp.AddTool() but does NOT include InputSchema field. Schemas provided in evaluation metadata appear inferred, not explicitly defined in the codebase.

get_cluster_statuslist_podslist_serviceslist_deploymentslist_nodeslist_namespacesget_resourceget_resource_yamlget_eventsget_pod_logscheck_rbac_permissionlist_configmapslist_statefulsets

HIGHpattern:tool-description ↗

Parameter descriptions lack actionable constraints. Examples: 'Kubernetes namespace name (required)' does not specify valid format, length limits, or character restrictions. 'Kubernetes resource type' does not document allowed values (pods, services, deployments, etc.) as an enum.

list_podslist_serviceslist_deploymentsget_resourceget_resource_yamlget_eventsget_pod_logscheck_rbac_permission

Recommendations

Add explicit JSON Schema InputSchema to every tool registration in internal/mcp/server.go. For namespace parameters, use type: 'string', minLength: 1, maxLength: 63, pattern: '^[a-z0-9]([-a-z0-9]*[a-z0-9])?$' (RFC 1123 subdomain). For resource_type, declare an enum: ['pod', 'pods', 'service', 'services', 'deployment', 'deployments', 'statefulset', 'statefulsets', 'configmap', 'configmaps', 'secret', 'secrets', 'daemonset', 'daemonsets', 'node', 'nodes', 'namespace', 'namespaces', 'event', 'events'].
Document output schema for each tool. Example for list_pods: returns object with fields 'items' (array of Pod objects with name, namespace, status, restarts), 'total' (int), 'nextCursor' (optional string for pagination). Include field types and descriptions.
Add pagination to all list_* tools. Introduce parameters: limit (int, default 20, min 1, max 500), offset or cursor (string, optional). Return total count and next_cursor for navigation. Include pagination guidance in description: 'Use limit=20 for large clusters to avoid context window exhaustion.'
Implement readOnlyHint annotation on all 13 tools. Update tool registration to include Annotations: struct { ReadOnlyHint: true } for client safety.
Replace free-form parameter descriptions with constraint-aware text. For namespace param: 'Kubernetes namespace name (required, RFC 1123 subdomain format: lowercase alphanumeric and hyphens, max 63 chars)'. For container_name in get_pod_logs: 'Container name within the pod (optional; if omitted, defaults to first container).'

Spec posture evidence

Inferred effective spec: 2026-07-28+.

Relies on Logging (deprecated) - log to stderr or use OpenTelemetry

Score history

Overall score trend

↑ 13 points across a rubric change (v1 → v2)

47/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-22	F	47	2026-07-28+	v2
2026-03-09	F	34	-	v1