This MCP server has significant structural and documentation gaps. While 17 tools are registered with names and basic descriptions, most lack adequate parameter documentation, error handling guidance, and output schema clarity. Many tool descriptions are generic or incomplete, and several critical tools expose secrets as parameters (ak, sk) rather than using server-side injection. The schema quality is inconsistent: some tools have detailed input schemas (e.g., search_dockstore_workflows) while others are sparse (e.g., validate_wdl). Tool naming is reasonable (verb-noun convention mostly followed), but composition issues exist, multiple tools operate on credentials that should be handled via environment variables. Output documentation is nearly absent across all tools. Error handling is not evident in visible code. This server would benefit from systematic documentation, parameter refactoring to remove secrets, and output schema definition.

• Refactor ALL 12 tools with ak/sk parameters to inject credentials server-side via environment variables (MIRACLE_ACCESS_KEY, MIRACLE_SECRET_KEY). Remove ak, sk from tool input schemas. Update descriptions to state 'Uses Bio-OS credentials from server configuration.' This prevents credential leakage.
• Document output schema for every tool. For list_submissions and list_workflows: explicitly state return type is {submissions: [{id, name, status, ...}], total: number, next_page_token?: string}. For validate_wdl: state return is {valid: boolean, errors?: [{line, column, message}]} or equivalent. Make schemas visible to LLMs.
• Add error handling guidance to tool descriptions. Example for validate_wdl: 'On validation failure, returns detailed error messages with line numbers. If syntax is correct but logical issues exist, returns warnings. Retry is safe, this tool is read-only.'
• Add confirmation/dry-run support to delete_submission. Either: (a) add optional dry_run: true parameter to preview what will be deleted, or (b) return a confirmation response requiring user approval before deletion. Prevents accidental data loss.
• Enrich parameter descriptions with constraints and defaults. Example: 'workspace_name (required, 1-64 chars, alphanumeric + hyphens): Name of the Bio-OS workspace. Defaults to env var BIOOS_WORKSPACE if not provided.' This guides LLM input.
• Document pagination limits and result caps. For list_submissions and list_workflows: state 'Maximum page_size is 100. Results are capped at 50 items per page if not specified. Supports cursor-based pagination via next_page_token.' Prevents unbounded queries.