This server has 20 well-structured tools with explicit Zod schema definitions and clear descriptions. Naming follows verb_noun conventions (things_add_*, things_get_*, things_update_*) and tool purposes are unambiguous. Descriptions range 20 - 200 characters and explain WHAT each tool does and WHEN to use it. However, parameter descriptions are inconsistently detailed, many READ_ONLY tools have minimal descriptions ('Maximum number of results to return' lacks guidance on default or max), and complex tools like things_add_todo do not document parameter relationships or constraints. Output schemas are not explicitly documented in the codebase, and error handling, while present via handleError(), lacks actionable recovery guidance (e.g., 'User not found, try search_users()'). Composition is good: each tool has one responsibility, naming is consistent, and related tools are clearly distinguished (get_today vs get_upcoming). Security is adequate: no credentials exposed as parameters, validation via Zod. Missing: per-parameter format constraints, output schema documentation, idempotent operation guarantees, and composition across tools (e.g., how to chain add_todo with update_todo).
Add todos and headings to an existing project. Items are added as a flat array where headings act as visual separators for the todos that follow them.
Creates a project in Things with support for sections (headings) and todos. Perfect for complex projects with multiple phases, days, or categories. Headings act as visual dividers - todos that follow a heading in the items array will appear grouped under it.
Add a new to-do to Things with full support for checklists to break down complex tasks into manageable steps. Use checklist when task has multiple components that don't warrant a separate project.
Get all to-dos in Anytime
Get all items in a specific area
Get all areas
Simple READ-ONLY GET tools lack meaningful descriptions and parameter guidance. Tools like things_get_inbox, things_get_today, etc. have only 3 words of description ('Get all to-dos in the Inbox'). Parameter 'max_results' lacks explanation of default value, maximum allowed, or pagination behavior.
Output schemas are not documented. The codebase shows tool implementations but does not provide JSON Schema or field descriptions for what each tool returns. LLMs cannot reason about downstream tool calls or field extraction without knowing the response structure.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 56 | 2026-07-28+ | v2 |
| 2026-03-09 | D | 52 | - | v1 |
Get all to-dos in the Inbox
Get all to-dos from a specific list by name
Get all completed to-dos in the Logbook
Get all to-dos in a specific project
Get all active projects
Get all to-dos in Someday
Get all tags
Get all to-dos scheduled for Today
Get detailed information about a specific to-do including deadline, notes, status, etc.
Get all deleted to-dos in the Trash
Get all scheduled to-dos (with dates)
Navigate to a specific item or list in Things
Update an existing project in Things using JSON API for full feature support
Update an existing to-do in Things using JSON API for full feature support
Error handling does not provide recovery guidance. The handleError() method in AbstractToolHandler returns a generic error message (e.g., 'Unknown tool' or ZodError messages) without suggesting next steps. An LLM receiving 'Invalid parameters' has no guidance on what to correct or which tool to try instead.
Complex parameters lack constraint documentation. things_add_todo accepts 'when' and 'deadline' as free-form strings but provides no format guidance (ISO 8601? Natural language like 'tomorrow'?). things_add_project 'items' parameter is described as 'Items (todos and headings) to include' but does not specify the expected object structure or required fields.
Pagination is not documented. GET tools accept max_results but do not clarify whether results are paginated via offset/limit, cursor, or a single batch. No mention of whether there is a default limit or how to fetch additional pages, preventing LLMs from understanding how to retrieve large result sets.
No idempotency guarantees documented. WRITE tools like things_add_todo and things_update_todo do not state whether they are safe to retry. Agents may retry on ambiguous failures (network timeout, slow response), a non-idempotent tool risks duplicate todos, duplicate projects, or unintended side effects.
Tool composition not documented. things_add_todo and things_add_project succeed but do not state what fields (e.g., 'id') are returned, making it unclear whether an LLM can immediately chain to things_update_todo or things_add_items_to_project. Missing chaining IDs forces discovery detours.