Server has 6 tools with complete schemas and descriptions, but quality is uneven. All tools follow verb_noun naming (list_, get_, search_, save_, update_, delete_). Descriptions are present (avg ~90 chars) but lack actionable detail, they state WHAT but not WHEN or WHY. Parameters have types and descriptions, but several lack constraints (e.g., 'query' in search_bookmarks has no length/format guidance). No output schemas documented. Error handling is minimal, responses are JSON blobs without recovery guidance. Veyra token exposure as a parameter violates secret-injection pattern. No tool annotations (readOnlyHint, destructiveHint). Composition is sound: tools are single-purpose and chainable.

• Move veyra_token to server-side environment variable or vault. Inject it internally; never expose as a parameter. Update save_bookmark, update_bookmark, delete_bookmark to remove veyra_token param.
• Document output schemas for all tools. E.g., list_bookmarks returns {count: number, bookmarks: [{id, url, title, tags, category, created_at, updated_at}]}. Include in tool description or separate schema field.
• Add tool annotations: readOnlyHint=true for list_bookmarks, get_bookmark, search_bookmarks; destructiveHint=true for delete_bookmark. This signals agent safety constraints.
• Enhance descriptions with WHEN/WHY context. E.g., 'Search bookmarks by keyword (url, title, tags, category). Use this for fuzzy matching; use list_bookmarks to retrieve all or filter by exact tag/category.'
• Improve error responses with actionable recovery. E.g., if bookmark not found, return available IDs or suggest list_bookmarks. If Veyra token invalid, return 'Token expired or invalid. Regenerate at veyra.to.'.