Cover letter generation MCP server that produces PDF cover letters based on resume data and job descriptions using LaTeX
Single tool 'generate_cover_letter' has a complete input schema with proper types and descriptions for all 6 parameters, meeting baseline expectations (100% of params described). Tool name starts with verb 'generate' which is action-oriented. However, the tool description is minimal (52 chars: 'Produce a PDF file for the cover letter.'), falling below the 10-1024 char guideline sweet spot (ideal 10 - 200 chars for clarity). Output is undocumented, returns a plain string status message with no structured schema or field documentation, forcing LLMs to parse free-form text. No error categorization, recovery guidance, or actionable error messages are present (e.g., LaTeX compilation failures return only 'An error occurred during LaTeX compilation: ...' with no retry guidance). The tool performs a stateful write operation (creates PDF file, modifies filesystem) but the description does not explicitly state it modifies state, making it ambiguous to LLMs whether this is a safe read-only query. Parameter values like company_address and skills use nested arrays (List[Tuple[str, str]]) which are well-typed but lack format constraints or examples in descriptions. No input validation hints or examples guide LLM usage.
Produce a PDF file for the cover letter.
Tool description is minimal (52 chars) and lacks context for LLM selection. Should explain WHAT it does, WHEN to use it, and what the output contains. Current: 'Produce a PDF file for the cover letter.' Missing: clarification that it writes files, takes 5-10 seconds (pdflatex), and returns a status string, not a file path or binary.
Output schema is undocumented. Tool returns a plain string ('Successfully compiled LaTeX' or error message), but LLMs have no guidance on what to expect, how to parse the response, or what fields are present. A documented schema stating {type: 'object', properties: {status: {type: 'string'}, pdf_path?: {type: 'string'}, error?: {type: 'string'}}} would enable proper LLM reasoning.
Error responses lack actionable recovery guidance. LaTeX compilation failures return 'An error occurred during LaTeX compilation: {stdout} {stderr}', raw error output with no suggestion for what the LLM should do next (e.g., 'Retry with corrected input, or ask user for feedback'). Per pattern:recovery-guide, errors must tell the LLM what to do.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 55 | 2026-07-28+ | v2 |
Tool description does not explicitly state that this is a WRITE operation with irreversible side effects (creates file, calls subprocess, consumes disk/CPU). Per pattern:command-tool, destructive operations must be clearly marked so LLMs know the call is not idempotent or safe to retry casually.
No input validation or constraint guidance in parameter descriptions. 'company_address' accepts a List[str] with description 'the address of the hiring company/organization (two lines max)', but this constraint is narrative, not machine-enforced. Parameter should specify min/max array length (1 - 2 items, each string 1 - 200 chars) and note that validation happens on call, not in schema.
Parameter 'skills' uses List[Tuple[str, str]] which lacks format guidance. LLMs do not have a natural mapping for tuples in JSON. Should be clarified as 'array of objects with skill (string) and description (string) fields' with an example in the prompt or tool context, not the description itself.
No confirmation or dry-run mechanism for an irreversible file-writing operation. Per pattern:confirmation-request, agents should have a chance to preview or confirm before executing. Tool could offer a 'preview_mode: bool' parameter that generates a preview string instead of writing PDF.