StreamsHub MCP provides 10 well-structured diagnostic tools for Kafka/Strimzi clusters. All tools have explicit descriptions (averaging ~150 chars, within the 34-392 baseline), clear action-verb naming, and documented input schemas with parameter descriptions. However, output schemas are not documented in the source code provided, and error handling guidance is minimal. Tool annotations (readOnlyHint) are present, which aligns with current MCP spec patterns. The tool set demonstrates good composition (each performs one diagnostic task), but lacks pagination guidance and result-limiting documentation for endpoints that may return large datasets.

• Document the output schema for each tool. Define the structure of the diagnostic data returned: what fields are present (status, conditions, metrics, logs, events), their types (string, object, array), and nesting. Example: 'Returns {cluster_status: string, conditions: [{type, status, message}], metrics: {replication_lag_ms: int, ...}, logs: [{timestamp, pod, message}], events: [{...}]}'.
• Add result limiting guidance. Specify in each diagnostic tool description: 'Returns a maximum of 100 events, 50 pod logs, and 10 metrics per request. Use sinceMinutes or filters to narrow results.' This prevents context window overflow.
• Implement error response specifications. Document what errors each tool can return and what recovery steps are valid. Example: 'If cluster not found: User should verify cluster name with list_kafka_clusters tool. If permission denied: Ensure RBAC allows access to Kafka and KafkaConnect resources in the namespace.'
• Formalize mutually exclusive constraints in schema. Use oneOf or allOf JSON Schema keywords to enforce that rangeMinutes cannot be combined with startTime/endTime, rather than documenting it only in description text.
• Enhance namespace parameter documentation. Clarify: 'If omitted, searches all namespaces accessible to the MCP server. If access is restricted to a single namespace, that namespace is searched. Namespace filters improve performance and reduce noise.'
• Add chaining IDs to response documentation. Ensure responses include cluster_name, namespace, connector_name, or other IDs needed by downstream diagnostic tools to enable multi-step workflows without extra lookup calls.