zlib-mcp
An MCP server that lets AI agents search and download books from z-library using the user's own account credentials.
README
zlib-mcp
A stdio MCP server that gives any AI agent tool — Claude Code, Codex CLI, Cursor, Claude Desktop — the ability to search z-library and download books.
Bring your own account. There is no shared backend, no API key, no proxy: the server runs on your machine, talks straight to z-library, and uses your credentials and your quota.
Tools
| Tool | What it does | Needs credentials |
|---|---|---|
zlib_search |
Search by title / author / ISBN, with format, language and year filters | yes |
zlib_get_download_url |
Get a direct download link for one book (no file written) | yes |
zlib_download |
Download a book into a directory you configured | yes |
zlib_limits |
Check today's remaining download allowance | yes |
zlib_login |
One-time helper: exchange email + password for remix credentials | no |
zlib_download only appears once you set ZLIB_DOWNLOAD_DIR — an MCP server that can write files anywhere by default is not an acceptable default, so you have to name the directory yourself.
Requirements
- Node.js ≥ 20
- A z-library account
Setup in 5 minutes
1. Get your credentials
If you already know your remix_userid / remix_userkey, skip ahead. Otherwise add the server with just your email and password (see the config snippets below), then ask your agent to run zlib_login once and put the returned remix_id / remix_key into the config permanently.
You can also run it straight from a terminal:
ZLIB_EMAIL=you@example.com ZLIB_PASSWORD='…' npx zlib-mcp
2. Add the server to your client
Every client takes the same three things: the command npx, the argument zlib-mcp, and an env block.
<details open> <summary><b>Claude Code</b></summary>
claude mcp add zlib \
--env ZLIB_REMIX_ID=123456 \
--env ZLIB_REMIX_KEY=your_remix_userkey \
--env ZLIB_DOWNLOAD_DIR="$HOME/Downloads/books" \
-- npx -y zlib-mcp
Or edit ~/.claude.json / .mcp.json directly using the JSON below.
</details>
<details> <summary><b>Claude Desktop</b> — <code>claude_desktop_config.json</code></summary>
macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json
{
"mcpServers": {
"zlib": {
"command": "npx",
"args": ["-y", "zlib-mcp"],
"env": {
"ZLIB_REMIX_ID": "123456",
"ZLIB_REMIX_KEY": "your_remix_userkey",
"ZLIB_DOWNLOAD_DIR": "/Users/you/Downloads/books"
}
}
}
}
</details>
<details> <summary><b>Cursor</b> — <code>.cursor/mcp.json</code> or <code>~/.cursor/mcp.json</code></summary>
{
"mcpServers": {
"zlib": {
"command": "npx",
"args": ["-y", "zlib-mcp"],
"env": {
"ZLIB_REMIX_ID": "123456",
"ZLIB_REMIX_KEY": "your_remix_userkey",
"ZLIB_DOWNLOAD_DIR": "/Users/you/Downloads/books"
}
}
}
}
</details>
<details> <summary><b>Codex CLI</b> — <code>~/.codex/config.toml</code></summary>
[mcp_servers.zlib]
command = "npx"
args = ["-y", "zlib-mcp"]
[mcp_servers.zlib.env]
ZLIB_REMIX_ID = "123456"
ZLIB_REMIX_KEY = "your_remix_userkey"
ZLIB_DOWNLOAD_DIR = "/Users/you/Downloads/books"
</details>
3. Try it
Find me Kleppmann's Designing Data-Intensive Applications in epub, then download the first result.
Configuration
| Variable | Required | Default | Notes |
|---|---|---|---|
ZLIB_REMIX_ID |
one of two | — | Your remix_userid |
ZLIB_REMIX_KEY |
one of two | — | Your remix_userkey |
ZLIB_EMAIL |
one of two | — | Fallback: exchanged for remix credentials on first use |
ZLIB_PASSWORD |
one of two | — | Fallback, used with ZLIB_EMAIL |
ZLIB_HOST |
no | pkuedu.xyz |
Upstream mirror; change it if you get blocked |
ZLIB_DOWNLOAD_DIR |
no | (unset → zlib_download disabled) |
Where downloads are written |
ZLIB_MAX_DOWNLOAD_BYTES |
no | 524288000 (500 MB) |
Files above this need allow_large: true |
ZLIB_TIMEOUT_MS |
no | 20000 |
Per-request connect timeout |
ZLIB_CREDENTIAL_CACHE |
no | 1 |
Set 0 to never write ~/.zlib-mcp/credentials.json |
ZLIB_LOG_LEVEL |
no | info |
debug / info / warn / error / silent; all logs go to stderr |
Credential precedence
ZLIB_REMIX_ID+ZLIB_REMIX_KEY- Cached credentials from a previous
ZLIB_EMAILlogin (~/.zlib-mcp/credentials.json, mode600) ZLIB_EMAIL+ZLIB_PASSWORD→ logged in on first tool call, not at startup
The cache exists so a client restart does not trigger a fresh login every time — repeated logins are what makes z-library's anti-abuse system notice you. It stores only the remix id and key; your password is never written to disk, logged, or returned by any tool. On Windows the 600 mode is a no-op (the OS ignores POSIX permissions) — set ZLIB_CREDENTIAL_CACHE=0 if that matters to you.
If nothing is configured the server still starts and lists its tools; calling one returns instructions on what to set. It does not crash — a crashed MCP server just shows up as "unavailable" in most clients, with nothing to debug.
Troubleshooting
"Upstream host … appears to be blocking this request" — the mirror is behind an anti-bot wall. Set ZLIB_HOST to another one and restart the client. Known mirrors change often; 1lib.sk is currently blocked, pkuedu.xyz currently works. Anything that serves the same /eapi/* endpoints will do.
"z-library rejected the current credentials" — your remix key expired. Run zlib_login again and update the config. If you use the email/password fallback, delete ~/.zlib-mcp/credentials.json to force a fresh login.
"download quota reached" — free accounts get a small number of downloads per day. zlib_limits shows the counter; it resets on z-library's side at midnight UTC.
Nothing appears in the client — check the client's MCP log; this server writes all diagnostics to stderr. ZLIB_LOG_LEVEL=debug makes it chattier.
zlib_download is missing — you did not set ZLIB_DOWNLOAD_DIR. That is by design.
Development
pnpm install
pnpm check # format check → lint → typecheck → tests
pnpm build
To try an unreleased version straight from git, point your client's command/args at
npx / ["-y", "github:shiyi-0x7f/zlib-mcp"] — the prepare script builds it on install.
Tests never hit the real upstream — fetch is stubbed everywhere.
Legal
This tool only provides API access to your own z-library account. It hosts nothing, distributes nothing, and ships no copyrighted content. Making sure your use of it is lawful where you are is on you. Your account, your quota, your risk — an account banned for abuse is yours to lose.
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。