Model Context Protocol server for Audible library, wishlist, collections, metadata, and listening stats.
audible-mcp demonstrates solid tool definition quality with consistent naming conventions, comprehensive descriptions, and well-structured input schemas using Zod validation. All 13 tools follow verb_noun naming pattern (list_, get_, search_). Descriptions are substantive (average ~100 chars, well within 10-1024 range). Input parameters have types and descriptions. However, output schemas are NOT explicitly documented, responses are passed through createJsonToolResult() which wraps them in JSON without formal schema declarations. Error handling uses createErrorToolResult() but lacks recovery guidance or error classification (retryable vs fatal). No tool annotations present (readOnlyHint, etc.). Security is well-handled: no credentials in parameters, auth delegated to AudibleApi class. Overall definition quality is B+, strong fundamentals, minor gaps in output documentation and error recovery patterns.
Return local auth metadata for the configured Audible device registration.
Fetch catalog metadata for an Audible ASIN from the catalog products endpoint.
Fetch chapter information for an Audible ASIN.
Fetch content metadata for an Audible ASIN, including chapter information.
Fetch a single library item by ASIN, including progress and download-related fields.
Fetch aggregate listening stats. The default window is the current month plus the prior two months.
Output schemas are NOT documented. Tools return responses via createJsonToolResult() which wraps data in {content, structuredContent} but callers have no formal specification of what fields to expect in the returned object (e.g. what properties does a library item contain?). This forces LLMs to infer structure on the fly, risking field lookup errors and limiting downstream tool composition.
Error responses lack recovery guidance. createErrorToolResult() returns only the error message with isError=true. Per the recovery-guide pattern, errors should guide the LLM on next steps (e.g. 'Authentication failed. Call audible_validate_auth to refresh credentials.').
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | C | 63 | 2025-06-18+ | v2 |
| 2026-03-09 | C | 60 | - | v1 |
List items inside an Audible collection by collection id.
List Audible collections for the authenticated account.
List titles in the library that have started but are not yet finished.
List titles from the authenticated Audible library.
List titles in the authenticated Audible wishlist.
Search the authenticated library by title, ASIN, author, narrator, or series name using client-side filtering across the first few pages.
Validate signed-auth access by performing a lightweight library read and refresh if needed.
No tool annotations present. None of the 13 tools declare readOnlyHint (all are read-only per Risk field) or idempotentHint. This metadata helps LLMs reason about retry safety and side effects. Per tool-annotations pattern, all read-only tools should include readOnlyHint annotation.
Pagination is implemented inconsistently. audible_list_library, audible_list_wishlist, audible_search_library, and audible_list_in_progress_titles accept optional page/numResults parameters, but descriptions do not state the total count, whether cursor-based or offset-based pagination is used, or what happens if page exceeds available data. Per paginated-result pattern, all paginating tools should document total count and next/previous cursor/offset in output.
Generic empty input schema for some tools. audible_list_collections and audible_get_auth_status declare inputSchema: {} with no explanation of why they take no parameters. While correct, descriptions could hint that these are stateless discovery tools (e.g. 'Returns all collections for the authenticated account; no filtering available.').