A MCP server for learning about and interacting with a candidate (personal information, resume, job search preferences)
Single tool 'contact_candidate' has a clear action verb but critical gaps in schema documentation, parameter descriptions, and error handling. The schema is partially visible in code but parameters lack individual descriptions. No output schema documented. Tool description is minimal (32 chars, below the 34-char p10 baseline). The callback implementation is a stub that always returns success without validation, error handling, or actual email transmission logic. This is below-median quality for production use.
Contact the candidate with a message
Tool description is too brief (32 characters). At 32 chars, it falls below the p10 baseline of 34 chars and provides minimal context for LLM selection. Should include when to use it, what happens (sends email), and what to expect in return.
Input parameters lack individual descriptions. The schema shows 'return_email', 'subject', and 'message' but provides no per-parameter documentation. LLMs cannot infer that return_email is the *sender's* address vs the recipient. Pattern requires 100% of A+ tools have param descriptions.
Output schema is not documented. Callers do not know what fields to expect from the response. The CallToolResult returns a TextContent with a generic success message, but downstream tools or agents cannot rely on a predictable response structure.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 51 | 2026-07-28+ | v2 |
No error handling or validation. The callback accepts any string values without validating email format, subject length, or message content. No guidance to LLMs on what to do if validation fails or email transmission fails. Pattern requires recovery guidance and actionable error messages.
Tool implementation is a stub. The callback only prints to stdout and returns a fake success message. It does not actually send emails or interact with any backend service. This is unusable in production and the description does not warn users of its placeholder status.
Destructive operation (sends email) lacks confirmation step. Pattern 'confirmation-request' requires irreversible operations to support dry-run or confirmation. No dry_run parameter or callback to confirm before sending.
No tool annotations. Modern MCP (2026-07-28 spec) supports destructiveHint and idempotentHint annotations. This tool modifies state (sends email) and should be annotated with destructiveHint=true so LLMs treat it as high-risk and plan carefully.
Parameter naming lacks specificity. 'return_email' is ambiguous, does it mean the sender's reply-to address? The recipient? Description should clarify. Best practice: 'sender_email' or 'reply_to_email' with a description explaining the distinction.
No input validation or constraints. Email parameter should have a regex pattern or description stating valid email format. Subject and message have no length limits. LLMs will hallucinate invalid inputs without explicit constraints.