Loading ToolBench
An MCP server for managing a parked-items queue in Claude Code. Provides interactive picking and deletion of parked tasks via native terminal UI.
Both tools lack input parameter schemas entirely. Tool descriptions are present but generic (60-80 chars), providing minimal guidance on when/why to use them. No documented output schemas. The 'unpark' tool description mentions interactive behavior but omits what data is returned to Claude. 'discard' warns about irreversibility but provides no guidance on recovery or confirmation flow. No parameter descriptions exist because no parameters are formally defined in the schema. Error handling is not documented. This server prioritizes UX (native pickers) over agent-facing clarity.
Open a native multi-select picker to delete one or more parked items from the queue. Check the items you want to remove, then press Enter to confirm deletion. Deleted items cannot be recovered.
Open a native picker to browse your parked items and pull one back into the conversation as your next task. An interactive list opens right in Claude Code (press the right arrow to expand it, up/down to choose, Enter to confirm); nothing is sent to the model until you pick. An empty queue or a read error shows a clear message instead of a blank list.
No input schemas defined for either tool. Both tools are registered via FastMCP but lack Pydantic models or JSON Schema definitions.
Tool descriptions are too brief (60 - 80 chars) and lack actionable context. 'Open a native picker to browse your parked items' does not explain what the tool returns, when to call it instead of alternatives, or what happens on error. Descriptions should be 50 - 200 chars with clear intent.
No output schema documented. LLMs cannot plan downstream actions without knowing what fields 'unpark' returns (e.g., does it return the full task text, an ID, or just a confirmation?). 'discard' provides no feedback on which items were deleted or why deletion failed.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 38 | 2026-07-28+ | v2 |
No error handling guidance. If the queue is empty, locked, or corrupted, the tool descriptions do not explain what the LLM should do next. 'discard' warns items 'cannot be recovered' but offers no undo tool or confirmation flow.
Tool names are generic action verbs without context. 'unpark' and 'discard' do not indicate they operate on a task queue. Better names: 'pull_parked_task' and 'delete_parked_tasks' would clarify the domain and reduce ambiguity.