This Java/Spring-based MCP server for OpenShift provides 25 tools with reasonable naming conventions (verb_noun pattern, action-oriented). However, the evaluation is severely constrained by incomplete source visibility, only one Java file is provided (OcpBuildTools.java, partially truncated), and the tool registration mechanism, output schemas, and error handling are not visible in the submitted code. Tool descriptions are present but often terse (10-50 chars), and no explicit output schema documentation is visible. The @ReactiveTool annotation suggests tool registration, but without seeing the full implementation across all 25 tool files, parameter validation, error handling, and output shaping cannot be assessed. Conservative scoring applies due to incomplete evidence.
Output schemas not documented. No visible tool return type definitions, LLMs cannot infer what fields to expect from responses, complicating downstream tool composition and field extraction.
Tool descriptions are terse (most 30-60 chars). Pattern baseline is 194 chars for production tools. Current descriptions lack WHEN to use context, prerequisites, or LLM selection guidance.
Expand tool descriptions to 100-200 characters. Include WHEN to call, prerequisites, and differentiation from similar tools. Example: 'ocp_get_cluster_version' should be 'Retrieves the OpenShift cluster version (e.g. 4.12.5). Call this first to verify cluster compatibility before attempting deployment operations.'
Document output schemas for all tools. Provide example responses with field names, types, and descriptions. List tools should document the structure of returned items (e.g. 'Returns array of ConfigMap objects with name, namespace, data fields').
Add pagination support to all list tools: accept limit (default 20, max 100) and offset/cursor parameters; return total_count and next_cursor in responses.
Implement per-tool error handling: categorize errors as retryable (network timeout, rate limit) vs user-fixable (resource not found, invalid parameter) vs fatal (permission denied, malformed request). Return specific recovery hints like 'Resource not found in namespace default. Available namespaces: kube-system, openshift-operators.'
Add tool annotations via MCP schema extensions: mark read-only tools with readOnlyHint=true (e.g. ocp_list_*, ocp_get_*), destructive tools with destructiveHint=true (ocp_delete_*), and idempotent tools with idempotentHint=true (ocp_restart_deployment).
Document namespace parameter behavior: specify whether it defaults to KUBECONFIG context, 'default', or requires explicit specification. Add note like 'If omitted, uses the active namespace from kubeconfig.'
Score history
Overall score trend
↑ 11 points across a rubric change (v1 → v2)
44/100
Scored
Grade
Overall
Spec posture
Rubric
2026-09-22
F
44
2026-07-28+
v2
2026-03-09
F
33
-
v1
Retrieves details of an ImageStream (tags, associated images)
No visible error handling or recovery guidance. Stack traces or raw error codes will not help LLMs self-correct. Pattern requires categorization (retryable, user-fixable, fatal) and actionable next steps.
Optional namespace parameter lacks guidance. LLMs will not know whether to infer a default namespace, ask the user, or use KUBECONFIG context. Describe fallback behavior.
No pagination parameters visible on list tools. Pattern baseline requires page/offset, limit, and total count for large result sets. Without pagination, list responses could exhaust context windows.
Provide confirmation/dry-run flow for destructive operations. Offer a separate tool (e.g. ocp_delete_configmap_dry_run) or add a dry_run boolean parameter to deletion tools.
Add composition guidance in descriptions: for create_configmap, note that the returned ConfigMap name can be passed directly to update_configmap or get_configmap. For trigger_build, specify what fields the returned Build object contains for polling completion.
Validate input formats at tool entry: enforce namespace naming (RFC 1123), ConfigMap data must be valid JSON, replicas must be non-negative integers. Return clear validation errors: 'Invalid namespace "my_namespace": must be lowercase alphanumeric and hyphens only, 1-63 chars.'
Include related IDs in responses to enable chaining. When listing Builds, return buildConfigName and buildId so agents can immediately call log/status tools. When getting a Node, include node labels/selectors for filtering.
Reduce optional parameters by using smart defaults. Default namespace to current KUBECONFIG context, default limit to 20 for list tools. This simplifies agent reasoning for common cases.
Provide visibility into source code for all 25 tools. Submit full implementations of OcpConfigMapTools, OcpDeploymentTools, OcpEventTools, OcpImageStreamTools, OcpJobTools, OcpNodeTools, and OcpClusterTools to verify error handling, validation, and output shaping across the suite.