An offline Model Context Protocol (MCP) server connecting local LLMs (LM Studio, Claude Desktop) to Kiwix Wikipedia (.zim) archives.

Offline Wikipedia MCP Server ๐Ÿ“š๐Ÿ”Œ

License: MITPython 3.10+MCP SDK

A fast, completely offline Model Context Protocol (MCP) server that connects local or cloud LLMs to locally downloaded Kiwix archives (.zim files).

This project bridges the gap between Large Language Models (running locally in LM Studio, Claude Desktop, or Cursor) and massive offline knowledge bases. By using the official Python MCP SDK and libzim, your LLM can search and read full Wikipedia articles โ€” and any other Kiwix archive โ€” without an internet connection.

โœจ Features

  • ๐Ÿ” Smart search โ€” uses full-text index search with automatic fallback to title/prefix search
  • ๐Ÿ“š Multi-archive support โ€” point at a folder and every .zim file inside is loaded automatically
  • ๐Ÿ“„ Direct article lookup โ€” retrieve a specific article by exact title, bypassing search
  • ๐Ÿท๏ธ Source labelling โ€” results show which archive they came from (e.g. [wikipedia_en_wp1-0.8_nopic_2026-04])
  • ๐Ÿงน Clean output โ€” strips HTML, scripts, tables, citation numbers [1], and [edit] markers
  • โœ‚๏ธ Configurable truncation โ€” prevent context-window overflows with WIKI_MAX_CHARS
  • ๐Ÿ”’ Thread-safe โ€” double-checked locking for safe use in concurrent MCP environments
  • ๐Ÿ“ฆ Any .zim file โ€” works with Wikipedia, Wiktionary, Stack Overflow, and other Kiwix archives

๐Ÿ› ๏ธ Tools Exposed to the LLM

Tool Description
search_wikipedia(query, num_results) Search all loaded archives using full-text index with fallback to title/prefix search
get_article(title) Fetch a specific article by its exact title, searching across all loaded archives

๐Ÿ“ฅ Step 1 โ€” Download Kiwix .zim Archives

Because LLMs only process text, it is highly recommended to download "nopic" (no pictures) versions to save disk space.

  1. Go to the Kiwix Library: https://library.kiwix.org/?lang=eng&q=wikipedia
  2. Recommended files:
    • Wikipedia English (wp1/Mini, nopic) โ€” most important articles, no images (~2 GB)
    • Wikipedia English (Maxi, nopic) โ€” full Wikipedia, no images (~22 GB)
  3. Save all .zim files into a single dedicated folder (e.g. ~/kiwix-zim/). The server will load every .zim file in that folder automatically.

โš ๏ธ macOS users: Do not store your .zim files inside ~/Documents, ~/Desktop, or ~/Downloads. macOS restricts app access to those folders and will cause an Error opening ZIM file. Use your home directory or a custom folder instead (e.g. ~/kiwix-zim/).

๐Ÿš€ Step 2 โ€” Installation

git clone https://github.com/Rome-NotBeepBoop/mcp-offline-wikipedia.git
cd mcp-offline-wikipedia

# Create and activate a virtual environment
python -m venv venv
source venv/bin/activate        # Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

โš™๏ธ Step 3 โ€” Configure Your MCP Client

Replace the example paths below with your actual paths.

โš ๏ธ Always use the full path to the venv Python (venv/bin/python), not the system python3. This guarantees the correct libzim version is used.

Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "offline-wikipedia": {
      "command": "/absolute/path/to/mcp-offline-wikipedia/venv/bin/python",
      "args": ["/absolute/path/to/mcp-offline-wikipedia/mcp_wiki_server.py"],
      "env": {
        "WIKI_ZIM_DIR": "/absolute/path/to/your/kiwix-zim/",
        "WIKI_MAX_CHARS": "15000"
      }
    }
  }
}

Cursor

Add the following to your Cursor MCP config (.cursor/mcp.json):

{
  "mcpServers": {
    "offline-wikipedia": {
      "command": "/absolute/path/to/mcp-offline-wikipedia/venv/bin/python",
      "args": ["/absolute/path/to/mcp-offline-wikipedia/mcp_wiki_server.py"],
      "env": {
        "WIKI_ZIM_DIR": "/absolute/path/to/your/kiwix-zim/"
      }
    }
  }
}

LM Studio

In LM Studio, go to Developer โ†’ mcp.json and fill in:

{
  "mcpServers": {
    "offline-wikipedia": {
      "command": "/absolute/path/to/mcp-offline-wikipedia/venv/bin/python",
      "args": ["/absolute/path/to/mcp-offline-wikipedia/mcp_wiki_server.py"],
      "env": {
        "WIKI_ZIM_DIR": "/absolute/path/to/your/kiwix-zim/",
        "WIKI_MAX_CHARS": "15000"
      }
    }
  }
}

๐ŸŒ Environment Variables

Variable Required Default Description
WIKI_ZIM_DIR โœ… Preferred โ€” Path to a folder โ€” every .zim file inside is loaded
WIKI_ZIM_PATH Legacy โ€” Path to a single .zim file (use WIKI_ZIM_DIR instead)
WIKI_MAX_CHARS No 15000 Max characters returned per article

WIKI_ZIM_DIR takes priority over WIKI_ZIM_PATH if both are set.

โ“ Troubleshooting

Neither WIKI_ZIM_DIR nor WIKI_ZIM_PATH is setโ†’ Make sure the env block is present in your MCP client config with the correct variable name.

ZIM archive not found at: ...โ†’ Double-check the path. On Windows, use forward slashes or escaped backslashes: C:/Users/you/kiwix-zim/.

macOS: Error opening ZIM file even though the file exists and the path is correctโ†’ macOS restricts app access to ~/Documents, ~/Desktop, and ~/Downloads without explicit permission. LM Studio (and the Python process it spawns) may be silently blocked. Two fixes:

  • Quick fix: Move your .zim files to ~/kiwix-zim/ or any other folder you created yourself, and update WIKI_ZIM_DIR accordingly.
  • Proper fix: Go to System Settings โ†’ Privacy & Security โ†’ Files and Folders and grant LM Studio access to the folder containing your .zim files.

Error opening ZIM file โ€” but the file opens fine in a Python terminalโ†’ LM Studio is using a different Python than your terminal. Always use the venv Python in your MCP config (venv/bin/python), not the system python3. Verify it has the right libzim:

/absolute/path/to/venv/bin/python -c "import libzim; print('ok')"

Server starts but the LLM says it can't find any articlesโ†’ Some .zim files don't have a full-text index. The server will automatically fall back to title search, but results may be less accurate for vague queries. Try using the exact article title.

High memory usageโ†’ Each .zim archive is loaded on first use. Larger archives (Maxi) will use more RAM. The Mini/nopic version is recommended for most use cases.

๐Ÿค Contributing

Contributions are welcome! Please open an issue first to discuss what you'd like to change. Pull requests should target the main branch.

  1. Fork the repo
  2. Create a feature branch (git checkout -b feature/my-improvement)
  3. Commit your changes (git commit -m 'Add some improvement')
  4. Push to your branch (git push origin feature/my-improvement)
  5. Open a Pull Request

๐Ÿ“œ License

This project is licensed under the MIT License โ€” see the LICENSE file for details.

MCP Server ยท Populars

MCP Server ยท New

    DeusData

    codebase-memory-mcp

    High-performance code intelligence MCP server. Indexes codebases into a persistent knowledge graph โ€” average repo in milliseconds. 159 languages, sub-ms queries, 99% fewer tokens. Single static binary, zero dependencies.

    Community DeusData
    joaoh82

    rustunnel

    Self-hosted, secure tunnel server in Rust. Expose local HTTP/HTTPS/TCP/UDP services to the public internet via TLS-encrypted WebSocket. Open-source, pay-as-you-go managed option, MCP server for AI agents.

    Community joaoh82
    GethosTheWalrus

    Temporal MCP Server

    MCP Server

    Community GethosTheWalrus
    SikamikanikoBG

    ๐Ÿ›ฐ๏ธ HomeLab Monitor

    Plug-and-play homelab dashboard in one container โ€” GPU, local-AI VRAM, Docker, systemd, host health. Built-in read-only MCP server so AI agents can explore it too.

    Community SikamikanikoBG
    w1ckedxt

    Cynical Sally

    Brutally honest senior-engineer code reviews for Claude Code, Cursor & Windsurf - and your terminal. Scores, evidence-backed issues, usable fixes.

    Community w1ckedxt