A Model Context Protocol server that generates ATS-optimized resumes using OpenAI API, LaTeX compilation, and AWS S3 storage.
Two tools with partial schemas and inconsistent documentation. generate_resume has a detailed input schema (form_data with nested objects, job_description, template_style) but lacks output schema documentation and error handling guidance. health_check has empty input schema and minimal description (27 chars, below 50-char baseline). Neither tool declares what happens on failure or how to recover. Tool names follow verb_noun pattern (good), but descriptions lack WHEN/WHY context. No parameter descriptions for template_style enum values. Missing idempotency and permission declarations.
Generate an ATS-optimized resume PDF from user data and job description. This tool: 1. Takes user form data and job description 2. Uses OpenAI to generate tailored LaTeX content 3. Compiles LaTeX to PDF using TinyTeX 4. Uploads PDF to S3 and returns presigned URL
Perform a health check on all system components. Returns: System health status and dependency information
health_check has empty input schema ({}). Description is 27 chars ('Perform a health check on all system components.\n\nReturns:\n System health status and dependency information'), below 50-char baseline and lacks WHEN/WHY context.
generate_resume lacks output schema documentation. Code shows ResumeResponse.dict() returned, but no schema is visible in tool definition. LLM cannot plan downstream calls or extract pdf_url, resume_id, expires_at without guessing structure.
template_style parameter has no description of valid enum values ('modern', 'professional', 'creative'). LLM must infer from code.
Error handling returns ErrorResponse with error_code and step fields, but no recovery guidance. 'Resume generation failed' tells LLM nothing about whether to retry, ask user, or abort. Per pattern:recovery-guide, errors must guide next action.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 44 | <=2025-11-25 | v2 |
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). generate_resume is destructive (writes to S3, calls OpenAI) but unmarked. health_check is read-only but unmarked. Per current spec (2026-07-28), annotations enable safer agent planning.