mcp-server-wikipedia 📚

This project exposes Wikipedia as an MCP server using a Progressive Retrieval Strategy. It is designed to minimize token usage by allowing LLMs to "scout" information before fetching large bodies of text.

_{Analytics powered by pypi.builditwithai.xyz}

The Problem: Token Waste

Wikipedia integrations often fetch multiple full pages up front, then decide what mattered. This fills the context window with irrelevant data and increases latency and cost.

The Solution: The Librarian Philosophy

This server implements a "Progressive Retrieval Ladder." Like a librarian helping you find a specific book, it encourages the model to:

1. Search for several candidate titles.
2. Summarize the candidates to find the right one.
3. Inspect the TOC to find the relevant section.
4. Fetch only the specific section OR the full page only if necessary.

graph TD
    A[Search Articles] --> B[Get Summaries]
    B --> C{Correct Page?}
    C -- No --> A
    C -- Yes --> D[Get TOC]
    D --> E[Get Section / Page]

Tools

• search_articles(query, limit=5): Top matching pages with snippets.
• get_summaries(titles): Compact summaries for multiple candidate pages.
• get_toc(title): Table of contents / section map for a page.
• get_section(title, section): Retrieve a single section by index or title.
• get_page(title): Retrieve the full plain-text page.

Token Efficiency Benchmark

In deterministic testing, this progressive strategy achieves up to 80% token reduction compared to naive full-page retrieval. Detailed results can be found in BENCHMARK.md.

Quick Start

Installation

From PyPI:

pip install mcp-server-wikipedia

Or run it directly via npx (if using the JS wrapper) or the python entry point:

python -m mcp_server_wikipedia

For development:

git clone https://github.com/surendranb/wikipedia-mcp-server.git
cd wikipedia-mcp-server
python3 -m venv .venv
source .venv/bin/activate
pip install -e .

Run

wikipedia-mcp-server

MCP Client Configuration

Claude Desktop

Add this to your claude_desktop_config.json:

{
  "mcpServers": {
    "wikipedia": {
      "command": "wikipedia-mcp-server"
    }
  }
}

Cursor / VS Code

Specify the wikipedia-mcp-server command in your MCP settings.

Example Prompts

• "Search for 'photosynthesis light dependent reactions' and summarize the top 3 candidates."
• "What molecules are produced during the light-dependent reactions of photosynthesis? Search first, then fetch only the relevant section."

Telemetry & Privacy

This server collects anonymous usage telemetry (like GA4 MCP and other sibling servers) so we can find and fix where the tools fall short: a random, resettable installation UUID, server version, OS, Python version, agent/client name, tool name, latency, result size, and error category. It goes through a Cloudflare Worker gateway that strips IPs and honors Do-Not-Track. No search queries, no page titles, no page content, no PII, no IPs — ever.

Opt out any time — any one of these works:

export WIKIPEDIA_MCP_TELEMETRY=false
# or
export DISABLE_TELEMETRY=1
# or
export DO_NOT_TRACK=1
# or
export NO_TELEMETRY=1

When opted out, nothing is sent and no local state is written (no ~/.wikipedia_mcp/ files, no process inspection). Delete ~/.wikipedia_mcp/ at any time to reset the anonymous installation id.

Development

Run tests:

python -m unittest discover -s tests -p "test_*.py" -v

Run benchmarks:

pip install -e ".[benchmark]"
python scripts/benchmark_token_efficiency.py

Contributing

We value simplicity and surgical efficiency. If you have an improvement that maintains the single-file architecture and enhances retrieval precision, we welcome your input. See CONTRIBUTING.md.

License

MIT License. See LICENSE for details.