The server has 2 tools with moderate quality. analyze_resume_gaps has a strong description (194 chars, well-structured) and detailed input schema (2 params, both with descriptions). However, the output schema relies on a Pydantic model (GapAnalysisResult) defined in prompts.py that is not fully visible in the provided code snippet, the schema structure cannot be fully verified. get_sample_resume is weaker: it's defined as a resource (not a tool), has a terse description (28 chars), and accepts only 1 parameter with minimal context. Both tools lack error guidance that tells LLMs what to do on failure (e.g., 'If the file format is unsupported, try converting to .docx first'). Parameter validation exists (file existence, format checks) but error messages are generic. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) are present. The naming convention is good (analyze_resume_gaps, get_sample_resume both start with action verbs), but the composition could be clearer, analyze_resume_gaps performs extraction, comparison, and recommendation in one call, which is acceptable for a complex domain operation but could benefit from clearer intermediate step guidance.

• Complete and expose the full GapAnalysisResult schema in the tool definition. Ensure all fields are documented with descriptions, types, and optionality (required vs optional). Verify it matches the expected output of LLM-friendly structured data (e.g., no verbose metadata, dates in ISO 8601, lists with counts).
• Enhance error messages with recovery guidance. For example: 'Resume file format not supported. Ensure the file is .docx format. If you have a .pdf or .doc file, try converting it to .docx first.' Instead of just 'ValueError: Only .docx files are supported'.
• Expand get_sample_resume description to 80-150 chars explaining: what sample resumes are available, how they're used (e.g., 'Returns the text content of a pre-loaded resume template for testing. Use this to understand the structure before providing your own resume.'), and a discovery hint: 'Call this to test analyze_resume_gaps before processing a real resume.'
• Add tool annotations: set readOnlyHint=true on both tools. This signals to LLM schedulers that these tools cannot modify state and can be called in any order.
• Document the 'name' parameter in get_sample_resume with an enum or constraint. Add a note: 'Available sample resume names: [list dynamically or examples]. Call list_sample_resumes() first to discover valid names.' (or embed the list in the error message when a sample is not found, as the current code does).
• Add a scope/permission declaration in the tool descriptions or server metadata: 'Requires OpenAI API access with gpt-4o-mini model capability. Ensure your OPENAI_API_KEY environment variable is set and has this model enabled.'