MCP server for Pinned — lets Claude Desktop, Cursor, Cline, and any MCP-aware AI tool call Pinned at the right lifecycle moments. Exposes pinned_before_code_change, pinned_before_done_check, pinned_scan_diff, pinned_list_guards, pinned_check_pr_description, pinned_suggest_init with structured responses agents must report in their final answer.
Six tools with READ_ONLY risk classification and basic descriptions, but significant gaps in schema documentation and parameter detail. Tool names follow verb_noun pattern (pinned_*) appropriately. Descriptions exist but are brief (avg ~80 chars) and lack actionable context for LLM selection. No visible input schema definitions in source code, parameters are inferred from brief descriptions only. No output schemas documented. Error handling and recovery guidance absent. Composition is reasonable (each tool has single responsibility), but parameter constraints and validation rules are not specified.
Called before code changes are made. Validates that the change is safe according to Pinned guards.
Called before marking a task as done. Checks if all Pinned guards pass.
Validates a PR description against Pinned requirements.
Lists all active Pinned guards in the repository.
Scans a code diff for potential issues and guard violations.
Suggests initial Pinned configuration for a repository.
No input schemas visible in source code. All six tools show only parameter names and brief descriptions; JSON Schema type definitions, constraints, enums, and validation rules are absent or not exposed.
No output schemas documented. Tools return structured responses (per package.json description) but response field names, types, and structure are not specified in visible code.
Descriptions are generic and lack actionable context. E.g., 'Validates that the change is safe' does not explain WHEN to call this vs pinned_scan_diff, what 'safe' means, or what the LLM should do with the result.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 41 | 2026-07-28+ | v2 |
No error handling or recovery guidance. Tools do not document what errors are possible, whether they are retryable, or what the LLM should do if validation fails.
Parameter descriptions are minimal. 'The code diff to validate' and 'The task context to validate' lack format guidance, size limits, or examples of what constitutes valid input.