zlib-mcp

zlib-mcp wraps Z-Library search and download operations as MCP tools. In an existing AI client, you can find books, compare editions, and then choose whether to request a link or save a file. The server runs locally over stdio and contacts the configured upstream with your own account. It does not provide shared credentials, a web reader, or full-text analysis. Its source code is MIT-licensed; access only material you are authorized to obtain.

Prepare Node.js 20+, add the README's npx launch configuration to your MCP client, and supply a complete remix ID/key pair. Omit the download directory for search and links only; enable it when you need local files.

Help me install zlib-mcp from https://github.com/shiyi-0x7f/zlib-mcp. First read README.md, README.zh-CN.md, package.json, src/config.ts, and src/credentials.ts, then confirm the Node.js 20+ requirement and target client. Use the repository's npx -y zlib-mcp configuration for local stdio MCP, preserving existing client settings. Prefer a complete ZLIB_REMIX_ID and ZLIB_REMIX_KEY pair; set ZLIB_DOWNLOAD_DIR only if I need downloads. Ask before requesting credentials, granting additional permissions, or overwriting files. Do not expose secrets in responses, logs, or version control, and do not select an unknown upstream yourself. After restarting, verify that 4 or 5 tools are present, then with my agreement check quota and search for a specified title. Confirm the exact item and quota cost before requesting a link or saving a file. If no real account is available or the upstream is unreachable, clearly record what remains unverified instead of treating local tests as a successful real download.

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

1. 1
   Prepare the account and runtime

   Ensure Node.js 20+ is available. Prepare both ZLIB_REMIX_ID and ZLIB_REMIX_KEY and confirm that you trust the upstream. Do not commit real credentials to a repository. If you want local files, choose a dedicated download directory.

2. 2
   Add the launch configuration to one client

   For Claude Code, copy the add command on this page and replace the credentials. For Claude Desktop, merge the JSON into ~/Library/Application Support/Claude/claude_desktop_config.json on macOS or %APPDATA%\Claude\claude_desktop_config.json on Windows. Cursor uses .cursor/mcp.json or ~/.cursor/mcp.json. Preserve existing mcpServers entries, replace the example download path with your own, and remove ZLIB_DOWNLOAD_DIR if local downloads are unnecessary.

3. 3
   Restart and check search tools

   Restart the selected client and check for 4 or 5 tools. Call zlib_limits first, then search for a specific title or author with zlib_search. Add extensions, languages, or year filters as needed and inspect the edition metadata and id/hash.

4. 4
   Retrieve a file after selecting an item

   Take both id and hash from the same search result. Use zlib_get_download_url for a link or zlib_download to save into the configured directory, only after accepting the quota cost. Check the returned path, filename, and byte count. Resolve account or upstream errors before retrying instead of repeatedly requesting downloads.

This assessment is based on the README, configuration and tool code, license, and commit history, supplemented by a source build and 82 mocked tests. Tools have distinct roles, local downloads require explicit configuration, and several clients have documented setup examples. The project is small, and practical use depends on its upstream, account status, and quotas. No real account was tested, so this review cannot establish live service reliability.