An MCP server that reviews your code the way a kernel maintainer would: bluntly, technically, without diplomacy.
Four tools with consistent naming (verb_noun pattern: review_*) and complete input schemas. All tools have descriptions (60-90 chars, within baseline 34-392 range). Parameters are typed and mostly described. However, output schemas are completely undocumented, LLMs cannot plan downstream calls or extract structured results. Error handling is absent: no guidance on retryability, user-fixable vs fatal errors, or recovery paths. No tool annotations (readOnlyHint present in metadata but not in schema). Descriptions lack WHEN/WHY context and dependency hints. Parameter descriptions are minimal (e.g., 'Review harshness' lacks explanation of what each enum value means). No pagination, no result limits documented, no field-level output structure.
Review code provided as a string, with optional language detection and context.
Review a git commit by reference (hash, HEAD, branch name).
Review a git diff provided as a string.
Review a file from the filesystem by path.
Output schemas completely undocumented. LLMs cannot determine what fields to expect from review results, preventing structured extraction and downstream tool chaining.
No error handling guidance. Responses lack recovery hints (e.g., 'file not found, check path' or 'git error, verify repo_path'). LLMs cannot self-correct on failures.
Parameter descriptions lack constraint details. 'harshness' enum values (gentle, standard, brutal, kernel-maintainer) are not explained, LLMs cannot reason about which to select.
Tool descriptions lack WHEN/WHY context. 'Review code provided as a string' does not explain when to use review_code vs review_file vs review_diff. LLMs must guess.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | <=2025-11-25 | v2 |
No result limits or pagination documented. If reviews are verbose, responses could exhaust context window. Baseline for A+ tools: cap results and offer pagination.