Loading ToolBench
MCP server that exposes project context (Git state, diffs, architecture) as resources and provides task management tools for AI agents to autonomously manage tasks across sessions
Three tools with minimal descriptions (10-30 chars, well below 50-200 char baseline for LLM optimization). Input schemas present but sparse: get_tasks has empty properties; add_task and complete_task have minimal param docs. No output schemas documented. No error handling guidance. No tool annotations (readOnlyHint/destructiveHint). Naming is verb-forward (get_, add_, complete_) but descriptions do not explain WHEN to use each tool or what they return. This server falls well below production baseline.
Add a new task to the session.
Mark a task as completed.
Get the current list of tasks for this session.
All tool descriptions are 10 - 30 characters; rubric requires 50 - 200 chars for LLM optimization. Descriptions do not explain WHEN to use each tool, what it returns, or dependencies.
No parameter descriptions. 'label' in add_task and 'id' in complete_task lack context. LLMs cannot infer whether 'id' is numeric, string, or UUID format.
No output schemas documented. LLMs do not know what fields get_tasks returns (task IDs, labels, timestamps?) or what add_task/complete_task return on success.
No error handling guidance. If complete_task fails (task not found, already completed), LLM receives no recovery hint. No actionable error messages.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 39 | 2026-07-28+ | v2 |
Missing tool annotations. add_task and complete_task are destructive (WRITE risk) but lack destructiveHint annotation. get_tasks is read-only but lacks readOnlyHint. Agents cannot infer safety properties.