zlibrary-mcp

# Z-Library MCP Server [](https://github.com/loganrooks/zlibrary-mcp/actions/workflows/ci.yml) [](https://www.npmjs.com/package/zlibrary-mcp) [](https://github.com/loganrooks/zlibrary-mcp/blob/master/LICENSE) A Model Context Protocol (MCP) server that gives AI assistants -- Claude Code, Claude Desktop, RooCode, Cline -- the ability to search Z-Library, download books, and extract document content for Retrieval-Augmented Generation (RAG) workflows. Built with a Node.js/TypeScript MCP frontend and a Python bridge backend for document processing. For what this project is — and deliberately isn't — see [VISION.md](VISION.md). ## Quick Start **Prerequisites:** Node.js 22+, Python 3.10+, and [UV](https://docs.astral.sh/uv/) (`curl -LsSf https://astral.sh/uv/install.sh | sh`) ```bash npm install -g zlibrary-mcp cd "$(npm root -g)/zlibrary-mcp" && bash setup-uv.sh # one-time Python environment setup ``` Then add to your MCP client config (Claude Code `.mcp.json`, Claude Desktop `claude_desktop_config.json`): ```json { "mcpServers": { "zlibrary": { "command": "zlibrary-mcp", "env": { "ZLIBRARY_EMAIL": "your-email@example.com", "ZLIBRARY_PASSWORD": "your-password" } } } } ``` Developing or contributing? Install [from source](#option-b-from-source-for-development) instead. ## Architecture Overview ```mermaid flowchart LR subgraph "MCP Client" A[Claude Desktop / Claude Code / RooCode] end subgraph "stdio Transport" B[Node.js MCP Server<br/>src/index.ts] end subgraph "HTTP Transport" SG[SuperGateway :8000] end subgraph "Python Bridge" C[python_bridge.py] D[lib/rag/ modules] end subgraph "External" E[Z-Library EAPI] end A -->|stdio| B A -->|SSE/HTTP| SG -->|stdio| B B -->|PythonShell| C C --> D C -->|httpx| E ``` - **Node.js/TypeScript MCP Server**: 13 tools registered via McpServer `server.tool()` API (MCP SDK 1.25+) - **Python Bridge**: Z-Library EAPI client (httpx) + document processing (lib/rag/ modules) - **EAPI Transport**: JSON API endpoints at `/eapi/` bypass Cloudflare browser challenges - **UV-Based Python Environment**: Project-local `.venv/` with `uv.lock` for reproducible builds - **Vendored Z-Library Fork**: Custom EAPI client for search and file downloads - **RAG Pipeline**: EPUB/PDF/TXT extraction with quality detection, output to files (not memory) ## Available MCP Tools (13 Total) ### Search Tools (7) | Tool | Description | |------|-------------| | `search_books` | Basic search by keyword with filters | | `full_text_search` | Search within book contents | | `search_by_term` | Conceptual navigation via terms | | `search_by_author` | Advanced author search | | `search_advanced` | Fuzzy match detection with separate exact/fuzzy results | | `search_multi_source` | Parallel search across multiple sources | | `get_recent_books` | Recently added books | ### Metadata Tools (1) | Tool | Description | |------|-------------| | `get_book_metadata` | Complete metadata extraction (terms, descriptions, ratings) | ### Collection Tools (1) | Tool | Description | |------|-------------| | `fetch_booklist` | Expert-curated collection contents | ### Download & Processing Tools (2) | Tool | Description | |------|-------------| | `download_book_to_file` | Download with optional file-based RAG bundle output | | `process_document_for_rag` | Extract a file-based RAG bundle from EPUB/PDF/TXT | ### Utility Tools (2) | Tool | Description | |------|-------------| | `get_download_limits` | Check daily download quota | | `get_download_history` | View recent downloads | For complete parameter documentation, types, and examples, see [API Reference](docs/api.md). ## Installation ### Option A: npm (recommended) **Prerequisites:** Node.js 22+, Python 3.10+, [UV](https://docs.astral.sh/uv/) ```bash npm install -g zlibrary-mcp ``` Then set up the Python environment inside the installed package (one-time): ```bash cd "$(npm root -g)/zlibrary-mcp" curl -LsSf https://astral.sh/uv/install.sh | sh # Install UV if needed bash setup-uv.sh ``` The package ships the complete Python bridge (`lib/`, the vendored `zlibrary/` fork, `pyproject.toml`, `uv.lock`), so no clone is needed. Since v1.3.0 the release workflow verifies the tag against `package.json` and publishes with provenance, so the registry version tracks GitHub releases; if `npm view zlibrary-mcp version` ever trails the [latest release](https://github.com/loganrooks/zlibrary-mcp/releases), that is a release-pipeline bug worth filing. ### Option B: From Source (for development) **Prerequisites:** Node.js 22+, Python 3.10+, [UV](https://docs.astral.sh/uv/) ```bash # 1. Install UV (one-time, if not already installed) curl -LsSf https://astral.sh/uv/install.sh | sh # 2. Clone and build git clone https://github.com/loganrooks/zlibrary-mcp.git cd zlibrary-mcp git lfs pull # Hydrates LFS-tracked test PDFs (don't run `git lfs install`; # it conflicts with the repo's Husky-managed hooks) bash setup-uv.sh # Creates .venv/ and installs Python dependencies npm install # Installs Node.js dependencies npm run build # Compiles TypeScript to dist/ ``` **MCP client configuration (stdio transport):** Claude Code (`.mcp.json` in your project root): ```json { "mcpServers": { "zlibrary": { "command": "node", "args": ["/absolute/path/to/zlibrary-mcp/dist/index.js"], "env": { "ZLIBRARY_EMAIL": "your-email@example.com", "ZLIBRARY_PASSWORD": "your-password" } } } } ``` RooCode / Cline (`mcp_settings.json`): ```json { "mcpServers": { "zlibrary-local": { "command": "node", "args": ["/absolute/path/to/zlibrary-mcp/dist/index.js"], "env": { "ZLIBRARY_EMAIL": "your-email@example.com", "ZLIBRARY_PASSWORD": "your-password" }, "transport": "stdio", "enabled": true } } } ``` ### Option C: Docker (HTTP transport) **Prerequisites:** Docker Versioned images are published to GHCR on every release (`latest`, `1.3`, `1.3.2`, …) — no clone or build needed: ```bash docker run -d --name zlibrary-mcp -p 8000:8000 \ -e ZLIBRARY_EMAIL="your-email@example.com" \ -e ZLIBRARY_PASSWORD="your-password" \ -v "$PWD/downloads:/app/downloads" \ ghcr.io/loganrooks/zlibrary-mcp:latest ``` The image wraps the stdio server in [SuperGateway](https://github.com/supercorp-ai/supergateway), exposing MCP over SSE at `http://localhost:8000/sse`. Verify it's serving: ```bash curl -s -N --max-time 3 http://localhost:8000/sse | head -2 # event: endpoint # data: /message?sessionId=... ``` > Alpine caveat: OpenCV has no musl wheels, so X-mark detection is unavailable > in the container; everything else works. **Building from source instead** (adds a `/health` endpoint via compose): ```bash git clone https://github.com/loganrooks/zlibrary-mcp.git && cd zlibrary-mcp cp docker/env.example docker/.env # then edit in your credentials docker compose -f docker/docker-compose.yaml up -d curl http://localhost:8000/health ``` **MCP client configuration (SSE/HTTP transport):** ```json { "mcpServers": { "zlibrary": { "command": "npx", "args": ["-y", "supergateway", "--sse", "http://localhost:8000/sse"] } } } ``` ## Output Format (RAG Processing) The RAG pipeline processes downloaded documents (EPUB, PDF, TXT) into clean text files for use in retrieval-augmented generation workflows. - **Output location:** Processed text files are saved to `./processed_rag_output/` - **File-based output:** Tools return file paths rather than raw text content, avoiding context overflow in AI assistants - **Bundle contract:** `processed_file_path` remains the body-text anchor, with additive sibling fields like `metadata_file_path`, optional `footnotes_file_path`, and an `output_files` map when available - **Supported formats:** EPUB, PDF, and TXT - **Quality detection:** The pipeline automatically detects document quality (OCR vs. digital text) and applies appropriate extraction strategies - **Scholarly formatting:** Preserves footnotes, chapter structure, and academic formatting where possible Use `download_book_to_file` with `process_for_rag: true` for combined download and processing, or `process_document_for_rag` to process an existing file. ## Configuration The server requires Z-Library credentials, set as environment variables: | Variable | Required | Description | |----------|----------|-------------| | `ZLIBRARY_EMAIL` | Yes | Z-Library account email | | `ZLIBRARY_PASSWORD` | Yes | Z-Library account password | | `ZLIBRARY_MIRROR` | No | Custom Z-Library mirror URL | The server validates credentials at startup and emits a clear error if they are missing. ### Logging Diagnostics go to **stderr**, never stdout: under the stdio transport stdout is the JSON-RPC channel, and anything else written there corrupts the protocol stream and causes clients to disconnect. Your MCP client captures stderr into its own logs. | Variable | Default | Description | |----------|---------|-------------| | `LOG_LEVEL` | `info` | `silent`, `error`, `warn`, `info`, or `debug` | `debug` adds per-request argument tracing, which **includes your search queries**. Scrub client logs before attaching them to a bug report. ### Diagnosing problems ```bash npm run doctor ``` This probes Z-Library, Anna's Archive, and LibGen directly and reports which respond. Those are undocumented third-party services that change domains without notice, so an upstream outage and a bug in this server look identical from an MCP client. Run this before filing an issue — and see [TROUBLESHOOTING.md](docs/TROUBLESHOOTING.md) for the setup-related causes of a server that exits immediately. ### Retry and Circuit Breaker All API operations include automatic retry with exponential backoff and circuit breaker protection. These are configurable via environment variables: | Variable | Default | Description | |----------|---------|-------------| | `RETRY_MAX_RETRIES` | `3` | Maximum retry attempts | | `RETRY_INITIAL_DELAY` | `1000` | Initial retry delay (ms) | | `RETRY_MAX_DELAY` | `30000` | Maximum retry delay (ms) | | `RETRY_FACTOR` | `2` | Exponential backoff multiplier | | `CIRCUIT_BREAKER_THRESHOLD` | `5` | Failures before opening circuit | | `CIRCUIT_BREAKER_TIMEOUT` | `60000` | Time (ms) before retry after circuit opens | See [docs/RETRY_CONFIGURATION.md](docs/RETRY_CONFIGURATION.md) for details. ## Development ### Running Tests ```bash # Node.js tests (Jest only -- npm test does not run pytest) npm test # Python tests -- fast suite (the routine command) uv run pytest -m "not slow and not integration and not performance" # Python tests -- full corpus (requires LFS-hydrated test PDFs) uv run pytest # Specific Python test uv run pytest __tests__/python/test_rag_processing.py ``` ### Building ```bash npm run build ``` ## FAQ ### Why EAPI instead of HTML scraping? Z-Library deployed Cloudflare browser challenges (Jan 2026) that block all HTML page requests from automated clients. The EAPI JSON endpoints at `/eapi/` are not subject to these challenges, providing reliable programmatic access. See `docs/adr/ADR-005-EAPI-Migration.md`. ### Why use `node` in .mcp.json instead of `uv`? The MCP server is a Node.js application. UV is only used during setup (`uv sync`) to create `.venv/` and install Python dependencies. At runtime, Node.js runs the MCP server which calls `.venv/bin/python` internally. ``` Setup: uv sync --> creates .venv/ Build: npm run build --> compiles TypeScript to dist/ Runtime: node dist/index.js --> runs MCP server --> calls .venv/bin/python internally ``` ### What if I move the project directory? No problem. `.venv/` moves with the project (UV creates project-local environments). ## Contributing Contributions are welcome! See [CONTRIBUTING.md](CONTRIBUTING.md) for setup instructions, code style, testing, and PR workflow. ## License This project is licensed under the MIT License - see the LICENSE file for details. ## Disclaimer This tool is provided for educational and research purposes only. Users are responsible for complying with all applicable laws and regulations regarding the downloading and use of copyrighted materials. Accessing Z-Library may be restricted in certain jurisdictions.