Public-domain ebook search aggregator exposed as an MCP server. Searches links only across built-in public-domain sources (Project Gutenberg, Open Library, Wikisource, Standard Ebooks, CBETA, etc.) plus user custom sources; downloads require explicit user selection and are restricted to built-in sources only.
Two tools with clear, detailed descriptions and well-structured schemas. Both tools have verb-noun naming (search, get) and comprehensive parameter documentation. However, output schemas are not formally documented in the code, and error handling lacks recovery guidance. The 'get' tool has strong security validation but descriptions could be more concise (currently 300+ chars, above the 200-char baseline for optimal LLM processing). Parameter descriptions are thorough but verbose.
Download one ebook file from a built-in public-domain source. Only URLs from the built-in sources are accepted: gutenberg.org, gutendex.com, standardebooks.org, openlibrary.org, archive.org, and raw.githubusercontent.com paths inside the wenshuoge or daizhigev20 public-domain repositories. Custom-source URLs are rejected - this tool never proxies user-configured sources. The downloaded file is verified by magic number before it is reported as success; failed files are deleted. 100 MB cap.
Search public-domain and open-license ebook sources. Fans out across all built-in sources (Project Gutenberg, Open Library / archive.org, Wikisource zh/en, Standard Ebooks, CBETA, wenshuoge, daizhigev20) plus the user's custom link-only sources, degrades gracefully when individual sources fail, and returns a ranked text report grouped by how the item can be obtained.
Output schemas not formally documented. Both tools return strings, but LLMs cannot parse structure without explicit schema documentation. 'search' returns a 'text report grouped by how the item can be obtained' but the structure is not defined. 'get' returns 'a short status message', unclear if it's JSON, plain text, or structured.
Error handling lacks recovery guidance. 'get' returns 'ERROR: {error}' strings but does not guide the LLM on next steps. E.g., 'host not in allowlist' should suggest 'Use search() to find valid sources' or 'Check custom source configuration'. Pattern: recovery-guide.
Tool descriptions exceed optimal length. 'search' is ~280 chars, 'get' is ~320 chars. Baseline for A+ tools is 50-200 chars. Verbose descriptions waste tokens and bury key details. Recommend: lead with action, defer implementation details to parameter descriptions.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 60 | 2026-07-28+ | v2 |
No pagination support. 'search' returns a 'ranked text report' but does not document result limits or pagination. If search returns 100+ results, the response could exhaust context. Recommend: cap results at 20-50 items and document the limit in the description.
Parameter 'language' in 'search' uses free-form string with default 'auto'. Should be an enum: ['zh', 'en', 'auto']. Current design invites hallucinated values like 'chinese' or 'english'.