Model Context Protocol server for LeetCode using GraphQL
This server exhibits significant gaps in tool definition quality. While all 7 tools have explicit registrations with Zod schemas and basic descriptions, the descriptions are often too terse (many under 20 characters for the tool itself), parameter annotations lack depth, and output schemas are completely undocumented. Error handling exists but is minimal (generic error wrapping without recovery guidance). The server follows a consistent structure across tools, but falls short of production-grade quality.
The URL slug of the contest
Get daily challenge
The URL slug of the problem (e.g., 'two-sum')
LeetCode username
LeetCode username
LeetCode username
Search problems
Tool descriptions are critically short and lack context. Six of seven tools have descriptions ≤20 characters (e.g., 'LeetCode username', 'Get daily challenge'), failing the 10-1024 character guideline. These descriptions do not convey WHAT the tool does, WHEN to use it, or what the return value contains.
Output schemas are completely undocumented. All tools return JSON.stringify(data) wrapped in a text content block, but there is no documentation of what 'data' contains, what fields are returned, or their types. LLMs cannot plan follow-up calls or extract specific values without knowing the response structure.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 46 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 38 | - | v1 |
Parameter descriptions lack specificity and format guidance. For example, 'contestSlug' and 'titleSlug' are described only as 'The URL slug of...' without explaining the expected format, examples (e.g., 'two-sum'), or how to obtain a valid slug. Parameter descriptions average ~35 characters vs. the 72-character baseline for A+ tools.
Error handling is generic and provides no recovery guidance. All tools catch errors and return 'Error: <message>' with isError=true, but do not classify errors as retryable/user-fixable/fatal, nor do they suggest next steps. An LLM encountering 'Error: 404 Not Found' has no actionable guidance.
No documentation of response field naming or structure. search-problems accepts 'limit' and 'skip' (pagination parameters) but no documentation of what the response includes, is there a 'total' count? 'next_cursor'? Are results an array or a nested object? This forces LLMs to reason about response structure on the fly.
Tool names are reasonably clear but descriptions fail to distinguish when to use similar tools. 'get-user-profile' and 'get-user-submissions' both require a 'username' parameter, but the descriptions do not clarify that profile returns metadata (rank, stats) while submissions returns code/attempts. An LLM may pick the wrong tool.