Model Context Protocol (MCP) server for IBM watsonx.data - enables AI assistants to query and explore lakehouse data
Server has 4 tools with explicit input schemas and descriptions. Naming follows verb_noun convention well (get_, list_, stop_, submit_). Descriptions are present and reasonably detailed (75 - 220 chars), though some lack guidance on when to use vs alternatives or actionable error recovery. Schemas are properly typed with object/string/array/integer, but parameter descriptions vary in specificity. Parameters like 'state' in list_spark_applications lack constraints (no enum for valid states). The submit_spark_application tool is complex with 11 parameters, some optional and interdependent (context_type, service_instance_id, type), but these dependencies are not well-documented. Output schemas are not explicitly documented, it's unclear what fields agents should expect back. No error handling patterns visible (no recovery guidance, no clarification on retryability). Security is handled via external IBM Cloud SDK (credentials not in params), which is good, but no audit logging or permission gates visible in tool definitions.
Get detailed status of a Spark application.
List Spark applications on a Spark engine.
Stop and remove a Spark application. This will terminate a running application and remove it from the engine's history.
Submit a Spark application for execution on a Spark engine.
Output schemas not documented. Tools define input schemas but do not specify what fields are returned, forcing LLMs to guess structure and making downstream tool composition risky.
No error handling guidance. Tools do not document retryability, user-fixable vs fatal errors, or recovery steps. E.g., stop_spark_application marked IRREVERSIBLE but no confirmation/dry-run pattern present.
Parameter constraints missing. list_spark_applications accepts 'state' as an open array of strings with no enum, agents could pass invalid states like ['unknown', 'pending']. Should define enum of valid states.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | C | 63 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 35 | - | v1 |
submit_spark_application has 11 parameters with undocumented interdependencies. 'context_type' (project|git_project|space) likely determines which other params are valid, but this is not stated. 'service_instance_id' expects 'iae'|'emr' with no enum. Parameter descriptions do not warn about mutually exclusive or conditional logic.
Pagination guidance incomplete. list_spark_applications accepts 'limit' (1 - 1000) but does not document offset/cursor behavior or total count in response. For large result sets, LLMs need to know how to iterate.
No confirmation step for destructive operations. stop_spark_application is marked IRREVERSIBLE (risk: IRREVERSIBLE) but tool definition does not offer a dry-run mode or explicit confirmation request pattern to prevent accidental termination.
Incomplete parameter descriptions. 'application' in submit_spark_application says 'Application file path (JAR, Python, R file)' but does not clarify: is this a local path, a remote URI, or a reference to a pre-uploaded artifact? Agents need exact formatting rules.