This server has 4 read-only tools with basic descriptions but significant quality gaps. All tools have empty input schemas (no parameters), which is appropriate for simple queries but limits expressiveness. Descriptions are adequate in length (120-160 chars) but lack critical contextual detail about when to use each tool versus alternatives, dependencies, and output structure. No parameter descriptions exist (because no parameters are defined). Output schemas are completely undocumented, LLMs don't know what fields to expect from each tool, making it hard to chain results. No error handling guidance. Tool naming follows verb_noun convention (get_*) which is good, but the four tools (get_active_tasks, get_all_tasks, get_active_projects, get_all_projects) have confusing similarity that creates LLM disambiguation friction. The tool set is minimal and read-only, which reduces security risk but also limits utility.

• Document the output schema for each tool in the tool description or via a separate structured annotation. For get_active_tasks, specify: 'Returns an array of task objects, each with: id (string), title (string), due_date (ISO 8601 string or null), priority (low|medium|high|none), project_id (string or null), notes (string), completed_date (ISO 8601 string or null). Limit: 500 tasks per call.'
• Clarify the distinction between 'active' and 'all' in tool descriptions. E.g., 'get_active_tasks returns tasks marked incomplete or on-hold; get_all_tasks includes completed tasks from the past 30 days. Use get_active_tasks for planning; use get_all_tasks only when the user explicitly requests historical completion data.'
• Add explicit guidance on error recovery in each tool description: 'If this tool fails, check that OmniFocus is running and the user has granted automation permission. Retry the tool after confirming. If it continues to fail, inform the user that OmniFocus automation is unavailable.'
• Consider renaming tools to reduce ambiguity: get_active_tasks → list_open_tasks, get_all_tasks → list_all_tasks (keep 'list' verb to imply returns a collection, and add an adjective like 'open' vs 'all' for clarity).
• Document pagination/limits: state the maximum number of items each tool returns and whether results are sorted (e.g., 'Returns up to 500 tasks, sorted by due date, oldest first.').
• Add per-tool metadata about side effects and idempotence in the description. E.g., 'This is a read-only tool. Calling it multiple times with the same input returns identical results and has no side effects.'