MCP server exposing Git Daily Review as agent tools (run reviews, read reports, curate the knowledge base)
Six tools with clear verb-noun naming and reasonable descriptions (avg 156 chars). All tools have input schemas with type definitions and parameter descriptions. However, output schemas are not documented in the visible code, parameter descriptions lack format/constraint details (e.g., date format YYYY-MM-DD is stated but not formalized), and error handling guidance is absent. No tool annotations (readOnlyHint/destructiveHint/idempotentHint) despite clear risk levels. Composition is sound, each tool has one responsibility and tools chain via IDs. Security concern: run_daily_review and decide_kb_suggestion are write operations but lack confirmation/dry-run patterns.
Approve (merge into the knowledge base) or reject a pending knowledge-base suggestion by its ID. Approval permanently updates config/knowledge-base.yaml.
Read the full Markdown daily report for a given date (findings, quality scores, critical issues, KB updates).
Show knowledge-base statistics: rule counts per layer, suggestion history, KB version.
List knowledge-base suggestions extracted from past reviews, optionally filtered by status (pending suggestions await human approval).
List the available daily reports (dates) and return the history index if present.
Run the AI code review over the day's commits of all configured repositories. Collects commits/diffs via git, reviews them against the project knowledge base, writes a Markdown report, and (optionally) extracts knowledge-base suggestions. May take several minutes depending on commit volume and the configured model.
Output schemas not documented. LLMs cannot infer what fields get_report, list_reports, list_kb_suggestions, and kb_stats return, forcing them to guess downstream field names and risk broken chains.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). run_daily_review and decide_kb_suggestion are write operations; LLMs should know these have side effects and are not safe to retry blindly.
No error handling guidance. If run_daily_review times out or decide_kb_suggestion rejects an invalid ID, the LLM has no recovery path. Errors should indicate retryability and next steps.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 69 | 2026-07-28+ | v2 |
Parameter descriptions lack formal constraints. 'date' is described as YYYY-MM-DD but not declared as a regex pattern or format. 'days' has min/max but no guidance on what happens if both 'date' and 'days' are provided.
No confirmation or dry-run for destructive operations. decide_kb_suggestion permanently updates config/knowledge-base.yaml; LLMs should have a way to preview changes or confirm before committing.