Z-Library → Mac → Kindle

Find a book, download it locally, and deliver it to the Kindle send-to address as an email attachment.

Delivery path: AgentMail (the Dross inbox, thedross@agentmail.to). AgentMail accepts base64 attachments, which is exactly what Amazon’s Send-to-Kindle address wants. Do not use raw SMTP for this — the send-kindle.py script in scripts/engagement/ does, and it works, but AgentMail is the requested path and gives a clean message-id receipt.

The script

Everything is driven by one script: {baseDir}/scripts/zlib_kindle.py

export PATH="/opt/homebrew/bin:$PATH"   # Homebrew python has the agentmail module
cd {baseDir}/scripts

Commands

Command	What it does
`python3 zlib_kindle.py search "<query>"`	List Z-Library matches with format, year, id, hash, URL
`python3 zlib_kindle.py search "<query>" --json`	Same, as JSON rows
`python3 zlib_kindle.py fetch <book-id> --hash <hash>`	Download one book to `~/Downloads/dross-books/`
`python3 zlib_kindle.py send <file> --title "..."`	Email a file to the Kindle address
`python3 zlib_kindle.py deliver "<query>"`	The whole pipeline — search → fetch → send
`python3 zlib_kindle.py account`	Check whether Z-Library credentials are configured
`python3 zlib_kindle.py books`	List curated titles with their Project Gutenberg ids

search prints the hash next to each id. Pass it to fetch with --hash — the eAPI resolves a book by the (id, hash) pair, so a bare id is not enough.

Typical flow

cd {baseDir}/scripts
python3 zlib_kindle.py search "The Fisherman John Langan" --limit 6
python3 zlib_kindle.py deliver "The Fisherman John Langan" --index 1

deliver picks the first result, downloads it, and emails it to amrekin@kindle.com. Use --index N to choose a different hit and --to <address> to override the destination.

Kindle prefers AZW3/MOBI. When the destination is a @kindle.com address, deliver looks for the same title in a Kindle-native format and switches to it automatically. Override with --format azw3|mobi|epub.

Trust the bytes, not the label. Z-Library’s advertised extension is often wrong — an “AZW3” row can serve MOBI bytes, and a MOBI downloaded as .txt is exactly what Amazon refuses to convert. sniff_ext() reads magic bytes and retarget_ext() renames the file to match, so check the printed filename after every download. MOBI/AZW3 are PalmDB containers: the BOOKMOBI/TPZ3 marker sits at byte 60, not byte 0.

Z-Library needs an account

Z-Library serves its file list only to signed-in users. Search works without an account, but resolving a download link does not: the eAPI /eapi/book/<id>/<hash>/file call only returns a link for a session that signed in. The script signs in on every run, before searching.

To enable downloads, put the account in:

~/.openclaw/workspace/secrets/zlib-account.txt

Format — one line, email:password (or email on line 1, password on line 2). The file is inside secrets/, which is git-ignored, so it will never be committed. secrets/zlib-account.txt is mode 600.

Check it with python3 zlib_kindle.py account — that command reports whether an account exists and never prints the password.

Without an account, the fetcher falls back to Project Gutenberg for public-domain titles, so the pipeline still completes end to end for those. In-copyright titles will fail at the download step with a clear error rather than silently producing nothing.

Project Gutenberg fallback

Titles in GUTENBERG_IDS map to numeric Gutenberg ids and are fetched directly from gutenberg.org. The script also queries gutendex.com to resolve ids it does not know. EPUB is preferred; the fetch verifies the file starts with the PK zip magic bytes before saving.

Books that are public domain

These have known Gutenberg ids and work without any Z-Library account:

The King in Yellow (8492), Carmilla (10007), The Call of Cthulhu (68283)
At the Mountains of Madness (70652), The Great God Pan (389)
The House of Souls (25016), The Turn of the Screw (209)
The Yellow Wallpaper (1952), The Picture of Dorian Gray (174)
Dracula (345), The Time Machine (35), Frankenstein (84)

Account limits — the two that bite

The Z-Library account has hard limits, and both are enforced server-side with a JSON body rather than an HTTP error code:

Daily download cap: 10 downloads/day. When it is reached the eAPI still answers success: 1 but withholds the download link, with allowDownload: false and a disallowDownloadMessage explaining the wait. The script prints that message.
Login rate limit. Every script run performs one sign-in, and too many in quick succession returns Too many logins #N. Try again later. Do not re-run the script in a loop to test it — one run is one login, and repeated probing trips the limit and blocks the real delivery as well.

When a run fails, read the stderr message before retrying. The script surfaces the server’s own explanation rather than a bare http 400.

Rules

Never print the Z-Library password or the AgentMail API key. The script reads both from disk and logs neither.
Use the Homebrew python (/opt/homebrew/bin/python3). System python has no agentmail module.
Verify the download before emailing: the script checks the zip magic bytes for EPUBs. Do not email a file that failed to download.
Report honestly. If a title could not be downloaded, say which step failed rather than claiming delivery.
A no-bounce check is the real delivery signal. SMTP sends do not appear in the IMAP Sent folder, so absence there proves nothing.
One script run is one Z-Library login. Do not loop the script to test it — repeated runs trip the login rate limit.

Notes on the site

The eAPI on z-lib.by is the working route. The public HTML pages now render a “Lifetime Premium” upsell instead of a download button, and their /papi/ endpoints 404. The eAPI is a JSON API: POST /eapi/user/login (email + password) → POST /eapi/book/search (message, page, limit, optional extensions[]) → GET /eapi/book/<id>/<hash>/file, which returns a signed CDN link.
eAPI titles arrive as "Author - Title". The script splits them back into title/author, and the author field often carries a bracketed duplicate (Hill, Joe [Hill, Joe]).
A book’s file is addressed by (id, hash), never by id alone.
The advertised extension and the served bytes can disagree — an “AZW3” row has been seen serving MOBI. Always let retarget_ext() have the last word on the filename.
The ego-browser skill exists but the binary is not installed, so browser-driven login is not available. It is not needed: the eAPI signs in with the stored credentials directly.