An MCP server that exposes a TODO list management service via the Model Context Protocol, integrated with Stytch authentication and backed by Cloudflare KV storage
This server defines 3 tools with basic naming conventions and descriptions. All tools start with action verbs (create_, mark_, delete_), which is good. Descriptions are concise but adequate (11-30 chars). Input schemas are present and typed with Zod, but parameter descriptions are minimal or missing. Output is formatted as structured text within a response object. The server lacks error handling guidance, parameter constraints (enums, ranges), and comprehensive schema documentation. Naming is clear but could be more specific (e.g., 'markTodoComplete' vs 'completeTodo'). No security annotations or idempotency hints are present.
Add a new TODO task
Mark a TODO as deleted
Mark a TODO as complete
Parameter descriptions are missing or trivial. The 'todoText' parameter in createTodo has a basic description 'The text content of the TODO task to create', but 'todoID' in markTodoComplete and deleteTodo lacks any constraint documentation (format, length, characters). Without descriptions, LLMs cannot validate input before calling.
No input validation constraints. Parameters accept free-form strings with no length limits, format patterns, or enums. A 'todoID' could be empty, null, or a malformed UUID without any client-side validation guidance. Error responses from TodoService failures are not visible in the MCP layer.
No error handling guidance. Tool implementations call this.todoService methods (add, markCompleted, delete) but do not show try-catch blocks, permission checks, or recovery hints. If a todoID does not exist, the LLM receives only the raw error state, not 'TODO not found. Try calling getTodos() first to list available IDs.'
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 55 | <=2025-11-25 | v2 |
| 2026-03-09 | D | 55 | - | v1 |
Output schema is undocumented. The formatResponse method returns a hardcoded structure with 'content' array containing a text field, but the MCP server definition does not declare this schema. LLMs cannot reason about what fields to extract from the response.
No tool annotations for destructive operations. The 'deleteTodo' tool modifies state irreversibly but lacks a destructiveHint annotation. The 'markTodoComplete' tool is listed as WRITE, 'deleteTodo' as DESTRUCTIVE, but no MCP-level annotations (e.g., tool.destructiveHint or tool.readOnlyHint) are visible in the server definition.
No confirmation or dry-run pattern for destructive operations. The deleteTodo tool does not support a 'confirm' parameter or separate 'preview' mode. Agents cannot verify intent before permanent deletion.
Missing resource integration for todo operations. The server defines a resource template for listing/reading todos (todoapp://todos/{id}), but there is no mechanism to tie tool output back to these resources. After createTodo succeeds, the LLM cannot easily reference the new todo via a URI.