This server has critical definition gaps across nearly all tools. Tool names are weak (verb-less or vague), descriptions are either missing or trivial, parameter descriptions are completely absent, and output schemas are undocumented. The server is explicitly STDIO-only, which caps its overall utility. Tool definitions are hardcoded in the ListToolsRequest handler, but lack the rigor expected for production agent use. Most tools expose parameters with no type constraints beyond raw JSON Schema; LLMs will struggle to know what values are valid. Error handling is minimal, no recovery guidance, no actionable error messages, and no confirmation/dry-run patterns for destructive operations like create-note.
Tools (5)
create-notewritesource verified55/100
Create a new Apple Note with specified title and content. Must be in HTML format WITHOUT newlines
get-noteread onlysource verified63/100
Get a note full content and details by title
index-notesread onlysource verified60/100
Index all my Apple Notes for Semantic Search. Please tell the user that the sync takes couple of seconds up to couple of minutes depending on how many notes you have.
Zero parameter descriptions across all tools. Parameters like 'title', 'query', 'content' are exposed with only JSON Schema types, no semantic guidance. An LLM cannot infer whether 'title' is a unique identifier, a substring to match, or an exact key lookup.
Output schemas are completely undocumented. Tools return data (list-notes returns string[], search-notes likely returns objects), but there is no specification of what fields are in each response. LLMs cannot reason about what data is available for downstream chaining.
create-note is a destructive tool with no confirmation/dry-run pattern. The description demands 'HTML format WITHOUT newlines' but does not explain how to handle user input that violates this constraint, nor does it offer a preview or confirmation step before committing the note.
Recommendations
Add a description to every parameter. For 'title' in get-note: 'The exact title of the note to retrieve. If the title is not found, return an error with suggestions for similar notes.' For 'query' in search-notes: 'A search string matching note titles or content. Supports partial matching. Will return up to 20 results sorted by relevance.'
Document all output schemas explicitly. For list-notes, state: 'Returns an array of note titles (strings). Example: ["Shopping List", "Meeting Notes - 2024-01-15"].' For search-notes: 'Returns an array of matching notes, each containing {title: string, preview: string (first 100 chars), creation_date: ISO8601}.'
Rewrite tool descriptions to follow the WHAT/WHEN/RETURN pattern. Current: 'Lists just the titles of all my Apple Notes'. Better: 'Retrieves all note titles in the user's Apple Notes. Use this first to discover available notes before fetching specific content. Returns an array of title strings.'
Add parameter type constraints. For create-note, specify: 'title: string, 1-255 characters. Must be unique; if a note with this title exists, the tool will return an error suggesting you update that note instead.' And 'content: string, valid HTML with no literal newlines. Use <br> tags for line breaks. If invalid HTML, the tool will return a validation error with the offending section.'
Implement dry-run support for create-note. Add an optional 'dry_run' parameter (boolean, default false). When true, validate the note but do not persist it. Return a preview of what will be created, allowing the agent to confirm before executing.
Tool descriptions are vague and under-specified. 'index-notes' says 'Please tell the user that the sync takes couple of seconds up to couple of minutes', this is instruction to the LLM that should not be in the tool description. Descriptions should state WHAT the tool does, WHEN to call it, and what it returns, not instruct the LLM on how to communicate with the user.
No error handling guidance. If get-note is called with a non-existent title, what error is returned? Is it retryable? Can the agent search for similar titles and retry? The code contains no recovery strategies or actionable error messages.
No parameter constraints or enums. For example, search-notes accepts 'query' as a free-form string, but provides no guidance on what constitutes a valid query. No length limits, no examples of query syntax, no specification of whether it matches titles, content, or both.
STDIO transport only. The server cannot be called remotely and is incompatible with hosted MCP clients or production agent deployments. This is a fundamental limitation.
Add error guidance for lookup failures. When get-note fails to find a note, return: 'Note titled "X" not found. Did you mean one of these? ["Meeting Notes - Jan", "Meeting Agenda"]'. This enables the agent to self-correct without an extra search call.
Index-notes description: remove the user-facing instruction. Replace 'Please tell the user that the sync takes couple of seconds up to couple of minutes...' with 'Indexes all notes for semantic search. Indexing may take 5 seconds to 2 minutes depending on library size. Returns when complete with a summary of indexed note count.'
Migrate from STDIO to Streamable HTTP transport (or at least HTTP+SSE) to enable remote client connectivity. STDIO is suitable only for local CLI tools, not production agents.
Add idempotency documentation. For create-note, specify: 'If called twice with identical title and content, the second call will return the existing note (idempotent behavior). If only title matches but content differs, this is an error, use update-note instead (if implemented).'
Return structured IDs and references that enable tool chaining. If search-notes returns note objects, include a 'note_id' (or similar) field so that get-note can accept either title or ID, reducing ambiguity when multiple notes have similar titles.