Six tools with complete input schemas and descriptions. Naming follows verb_noun pattern (get_, list_, add_, mark_, search_). Descriptions are adequate (50-150 chars) but lack dependency hints and recovery guidance. Parameters are typed with enums where appropriate (status filter). Error handling is sophisticated, typed errors map to user-facing text with context (cookie age, retry-after). However, output schemas are not documented in the tool definitions themselves; responses are inferred from code. No tool annotations (readOnlyHint/destructiveHint). Risk classification is present in metadata but not exposed to LLM.

• Add explicit output schema documentation to each tool. E.g., 'Returns: {items: [{id, title, author, kind, status}], total: number}' in the tool description or as a separate schema field.
• Annotate write tools (add_to_list, mark_finished) with destructiveHint=true and idempotentHint=false to signal to LLMs that these are state-changing and may have side effects.
• Enhance error descriptions with actionable recovery steps. E.g., 'Authentication failed. Call get_cookie_status to check cookie age, then update READSHELF_SESSION_COOKIE environment variable and restart the server.'
• Document pagination behavior in search_library: 'Results are limited to 50 items. If more exist, call search_library again with a refined query or different keywords.'
• Add a discovery tool or enhance get_cookie_status to return available operations and their prerequisites, guiding the LLM on tool sequencing.

Inferred effective spec: 2026-07-28+.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry