Loading ToolBench
MCP server for decidex — surfaces engineering decisions to Claude Code
Two tools with clear, actionable descriptions and well-structured input schemas. Both tools follow verb_noun naming (get_decisions, get_stats) and include parameter descriptions. However, output schemas are not documented in the visible code, and error handling guidance is absent. Descriptions are solid (120 - 150 chars, within the 10 - 1024 baseline) but lack dependency hints and recovery guidance. No tool annotations (readOnlyHint, idempotentHint) despite both being read-only operations. Parameter constraints are present (limit: 1 - 50) but could be more explicit in descriptions.
Retrieve engineering decisions relevant to a file or directory path. Call this when starting work on a file to surface decisions that apply to that area. Returns decisions sorted by recency, most recent first.
Get a summary of all captured engineering decisions — total count and breakdown by area. Useful for understanding how many decisions exist and which parts of the codebase have the most context.
Output schemas not documented. LLMs cannot infer what fields get_decisions and get_stats return, forcing them to guess downstream field names and risk broken tool chains.
No error handling guidance. Descriptions do not explain what happens if area path is invalid, if no decisions exist, or how to recover. LLMs receive no recovery hints.
Tool annotations missing. Both tools are read-only and idempotent, but lack readOnlyHint and idempotentHint annotations. Agents cannot distinguish safe tools from destructive ones.
Parameter description for 'area' lacks format guidance. Should specify: 'Relative path (e.g., src/auth/, src/api/users.ts). Use empty string for repo-wide decisions.' Current description omits the empty-string convention.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 62 | 2025-06-18+ | v2 |
No dependency hints. Description does not explain when to call get_stats first to understand decision coverage, or when get_decisions is the right choice vs. searching raw decision files.