The server exposes 11 tools with significant quality gaps. Only 3 tools (plan_workflow_with_tools, get_style_guide, retrieve_documents) have documented schemas visible in source. The remaining 8 tools (upload_model, add_class, add_attribute, add_connector, metadata_checker, reuse_check, style_guide_check, validator_check) are referenced but lack visible definitions, their schemas, parameter descriptions, and return types cannot be verified from the provided code. plan_workflow_with_tools has a complex schema with several parameters including a 'ctx' parameter of type 'object' that is poorly described ('Context object for accessing fastmcp server and sampling capabilities', vague and non-actionable). Tool descriptions range from good (plan_workflow_with_tools at ~250 chars) to minimal. get_style_guide has an empty input schema (no parameters documented). retrieve_documents accepts search_terms and vocabularies but lacks enum constraints on vocabularies. Error handling is minimal, no recovery guidance, no categorization of retry-ability. The codebase shows print-based debugging rather than structured logging. No evidence of input validation, permission gates, or security patterns. Tool naming follows verb-noun convention (plan_, get_, retrieve_, add_, validate_) which is positive, but composition issues exist: plan_workflow_with_tools is overloaded (it discovers tools, reasons, executes nested calls, and returns a plan, multiple concerns bundled). Missing: documented output schemas, pagination, error messages that guide recovery, parameter constraints (e.g. limit defaults and bounds), and idempotency markers.

• Complete all tool definitions in the server.py file with explicit schemas, descriptions, and return types. For each tool, include: a 50-200 character description answering WHAT it does, WHEN to use it, and what it returns; explicit input schema with types and descriptions for each parameter; documented return type (object or array with field descriptions).
• Add parameter constraints: retrieve_documents.limit should be documented as '1 - 1000, default 20' with explicit min/max. retrieve_documents.vocabularies should declare valid enum values like ['eGovernment', 'CoreVocabulary', 'ApplicationProfile'] to guide LLM selection.
• Split plan_workflow_with_tools into separate tools: 'plan_with_available_tools' (discovers and reasons about tools, returns a structured plan) and 'execute_plan' (executes a pre-built plan with specified tools). This respects single-responsibility and makes composition clearer.
• Replace 'ctx' parameter in plan_workflow_with_tools with explicit, typed parameters. Clarify what the LLM needs to pass (e.g., model_user, model_name) rather than an opaque context object.
• Document output schemas for all tools. Example for retrieve_documents: 'Returns array of {id, title, vocabulary, snippet, relevance_score}' so the LLM knows what fields to extract for follow-up calls.
• Add error handling and recovery guidance. Example: retrieve_documents should return 'No documents found for search_terms=[...]. Try: (1) simplify keywords, (2) expand vocabularies filter, (3) check spelling.' This lets LLMs self-correct without extra calls.