Multi-source ebook search across 10 legal sources (public-domain classics, CBETA/ctext Chinese canon, LibriVox audiobooks, community netdisk book-lists via full GitHub code search, Google Books metadata). Use when the user wants to find or download a book, ebook, EPUB/PDF/MOBI file, or audiobook in Chinese or English — even without the word find, any mention of a book title together with epub/pdf/下载/电子书 counts. 不适用: academic papers or journal articles (use a general web search tool). 触发词/Triggers: find a book, ebook search, epub, public domain books, audiobook, 找书, 电子书, 公版书, 有声书, 下载书.
Multi-source ebook search for legal public-domain, open-license, and
community-indexed books. The tool is a guide first: it surfaces ranked
links and direct download URLs; it only downloads after the user
explicitly asks for one.
IRON LAW:①绝不空手而归——公版源零命中不是终点,必须继续解析直达链接(微信读书 deepLink、GitHub 网盘书单、IA 借阅页),全部失败才降级到入口链接;把搜索页丢给用户自己去搜=任务失败。②下载必须等用户明确指令(「下载第N条」或同等表述)才执行,禁止预下载、禁止批量。③网盘结果一律 LINK_ONLY 并如实标注版权状态,工具不碰网盘,除非用户明确要求代取。
When to use
The user wants to find a book / ebook / EPUB / PDF / MOBI / audiobook.
Public-domain classics, ancient Chinese texts, the Buddhist canon.
Modern Chinese titles: community netdisk lists (github_lists source)
cover copyrighted bestsellers — present them as link-only hits.
Workflow
Run the search and wait for it to finish:
bash
bookscout search "书名"
Relay the results. The report groups hits by availability:
✅ direct download, 📖 read online / audiobook, 🔗 netdisk link-only,
🛒 borrow / purchase. Present each hit's number, source, URL and
formats in a clean readable list — summarize, do not dump raw output.
Zero hits does not end the task (IRON LAW ①). Try to resolve a
direct link before falling back to entry links:
For copyrighted Chinese books, open the WeRead search page
(https://weread.qq.com/web/search/global?keyword=...) and extract
the matching book's book-detail deep link to hand over directly.
Check the Internet Archive details link — borrowable with a free
account.
Only when neither resolves, relay the search-entry links the tool
printed (extended resources, manual entries, purchase channels).
Netdisk hits (ctfile / lanzou / baidu pan, marked link-only): give the
user the link plus its extraction code if present. If the user asks
you to fetch the file for them, follow
ctfile-playbook.md (ctfile links)
or the netdisk flow that matches the host — always on explicit user
request only (IRON LAW ③).
Stop and wait (IRON LAW ②). Only when the user explicitly says
"下载第N条" or an equally unambiguous confirmation, run:
bash
bookscout get N
Never download preemptively, never batch-download multiple items.
Environment (optional, unlocks more sources)
GITHUB_TOKEN / BOOKSCOUT_GITHUB_TOKEN — enables full GitHub code
search for netdisk book-lists (anonymous: curated seeds only).
BOOKSCOUT_GOOGLE_BOOKS_KEY — activates the Google Books source.
https_proxy / http_proxy — standard proxy vars; needed in some
networks to reach archive.org download nodes. See README Network FAQ.
Rules
Custom sources (user-configured) return link-only results. Give the
user the link; never call get for these hits.
If the search returns nothing, relay the manual entry points and
legitimate borrow/purchase channels printed by the tool. Never leave
the user empty-handed, and never invent sources the tool did not return.
Known limitations
[P] Anna's Archive / LibGen / archive.org download nodes unreachable
in some networks → blocked-network environments → hand over the
entry links, suggest the standard proxy env vars (README Network FAQ).
[P] Non-bestseller titles may have no netdisk backup anywhere
(book-list repos only index popular titles) → niche/professional
books → resolve a WeRead deep link plus purchase channels, say so
honestly instead of promising a file.
[P] Netdisk direct-link fetching needs a browser and breaks often
(anti-bot verification, site redesigns) → agent-assisted netdisk
downloads → fall back to link + extraction code for the user to
download manually.