MCP server for managing todos with support for multiple database adapters (in-memory and Cloudflare D1)
This is a well-structured todo MCP server with consistent naming, clear descriptions, and proper schema definitions across all 5 tools. All tools follow verb_noun naming (list_todos, add_todo, complete_todo, update_todo, delete_todo) and start with action verbs. Tool descriptions are concise (10-60 chars) and explain the action clearly. Input schemas are properly defined using Zod with type declarations and descriptions for all parameters. However, there are notable gaps: (1) output schemas are not formally documented, responses return unstructured text content rather than typed objects; (2) parameter descriptions could be more detailed (e.g., missing examples of what the 'done' state represents in filter options); (3) no error handling guidance for edge cases (e.g., what happens when deleting a non-existent todo); (4) missing per-tool output structure documentation that downstream tools could rely on; (5) no pagination or result limits documented despite list_todos potentially returning unbounded results. The server demonstrates solid foundational quality but lacks the polish and completeness expected of production-grade A-tier tools.
Add a new todo.
Mark a todo as complete.
Delete a todo.
Show all todos for the current user.
Update a todo.
Output schemas not formally documented. All tools return unstructured text content in a { content: [{ type: 'text', text: '...' }] } wrapper. Agents cannot reliably parse responses to extract todo IDs, completion states, or error details. This violates pattern:response-shaper and breaks downstream tool composition.
Destructive operation (delete_todo) lacks confirmation/dry-run support. No confirmation step or explicit warnings that this is irreversible. Agents can accidentally delete todos without safeguard. pattern:confirmation-request is not implemented.
Error messages are generic text responses with no classification or guidance. When delete_todo or complete_todo receives a non-existent ID, it returns plain text 'not found', no indication whether to retry, lookup the ID first, or abort. Violates pattern:recovery-guide and pattern:error-classification.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | C | 65 | 2026-07-28+ | v2 |
| 2026-03-09 | C | 64 | - | v1 |
Parameter descriptions contain awkward phrasing and lack format guidance. 'ID todo that about to be marked as complete' is grammatically incorrect and vague on ID format (UUID? numeric? slug?). Descriptions should state: 'The unique identifier of the todo to mark complete. Format: string'.
No pagination or result limits documented on list_todos. If a user has 1000 todos, the tool will return all of them as formatted text, potentially exhausting context window.
Tool compositions broken by missing return IDs. If add_todo returns only formatted text 'Todo #123 added: "..."', an agent cannot reliably extract the todo ID to pass to complete_todo or delete_todo. Response must return structured { id, title, done, created_at } object. Violates pattern:tool-chain.
update_todo schema does not support partial updates. Current schema requires both 'id' and 'title' parameters. If an agent wants to toggle done status without knowing the current title, it must first call list_todos to fetch the title, then call update_todo. No support for updating only the done flag. Design should allow independent updates or add a separate complete_todo parameter.
Idempotency not guaranteed. Adding the same todo twice with identical title creates duplicate entries (confirmed by handler logic). Agents may retry on timeout and create unwanted duplicates. Tools should be idempotent or explicitly document non-idempotent behavior. Missing: deduplication key or explicit idempotency guarantee.