A simple TODO list MCP server with in-memory task storage
Four tools with minimal descriptions (10-20 chars) and sparse parameter documentation. All tools have basic JSON Schema input definitions with type and description fields, but lack depth. No output schemas documented. Error handling is absent, delete and delete_all return void with no error guidance. Tool names are verb-based (add, all, delete, delete_all) but 'all' is vague (should be 'list_tasks'). No pagination, no idempotency hints, no recovery guidance. Schemas are present but descriptions fall below the 50-200 char baseline for LLM-optimized tools.
add new task
get all tasks
delete task by id
delete all tasks
All tool descriptions are under 20 characters ('add new task', 'get all tasks', 'delete task by id', 'delete all tasks'). LLMs cannot determine when/why to select tools with such minimal context.
delete() and delete_all() return void with no error handling, no confirmation step, and no recovery guidance. Destructive operations should support dry-run or require explicit confirmation to prevent accidental data loss.
Tool 'all' is vague, should be 'list_tasks' to follow verb_noun convention. 'all' does not clearly signal a read-only discovery operation.
No output schemas documented. Callers cannot infer what fields to expect from add(), all(), delete(), or delete_all(). add() and all() return structured JSON (Task and Task[]), but this is not declared in tool metadata.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 41 | <=2025-11-25 | v2 |
delete() and delete_all() have no return type annotation and no error classification. LLM cannot determine if operation succeeded, failed, or needs retry. No actionable error messages.