Wikipedia MCP Server

A Model Context Protocol (MCP) server that provides comprehensive Wikipedia access for AI assistants. Built in TypeScript, deployed on Vercel, and fully compatible with the MCP Streamable HTTP transport (protocol version 2025-03-26).

Features

• 🔍 Advanced Search — find articles with relevance ranking
• 📄 Full Article Access — plain text extracts or complete wikitext
• 📝 Smart Summaries — concise summaries tailored to a specific query
• 🗂️ Section Analysis — extract and summarize individual sections
• 🔗 Link Discovery — internal links and related articles
• 🌍 Multi-language — 50+ language codes, 140+ country codes, resolved automatically
• 🔀 Parallel Search — multi_search_wikipedia runs multiple queries simultaneously
• 🎯 Fact Extraction — structured key-fact lists from any article
• 📍 Coordinates — lat/lon for geographic articles
• ⚡ Streamable HTTP — full MCP transport support (POST + GET SSE + DELETE)
• 🔒 Security hardened — input sanitization, tiered rate limiting, no Server header
• 🚀 Vercel ready — stateless serverless deployment, zero cold-start friction

MCP Tools

All 13 tools accept optional language (e.g. "ja") and country (e.g. "JP") parameters. When set, the query is sent as-is to that language's Wikipedia endpoint (e.g. ja.wikipedia.org). Wikipedia's own search can often match a short English title phrase to the correct local article, but there is no translation layer in this server.

Tool reference

MCP Prompts

Three prompt templates are registered and available via prompts/list / prompts/get:

REST API

Standard HTTP endpoints for non-MCP usage:

GET  /health
GET  /search/:query?limit=10
GET  /article/:title
GET  /summary/:title
GET  /sections/:title
GET  /links/:title
GET  /coordinates/:title
GET  /related/:title?limit=10
GET  /summary/:title/query/:query/length/:maxLength
GET  /summary/:title/section/:section/length/:maxLength
GET  /facts/:title?topic=...&count=5
GET  /test-connectivity
GET  /supported-countries
POST /tools/:toolName

Architecture

src/
├── server.ts               # Express app wiring
├── routes/
│   ├── mcp.ts              # Streamable HTTP transport (POST + GET SSE + DELETE)
│   └── rest.ts             # REST endpoints
├── mcp-server.ts           # MCPServer helper — tool dispatch, language routing
├── handlers.ts             # Tool handler functions
├── toolRegistrations.ts    # SDK tool + prompt registration
├── toolDefinitions.ts      # JSON Schema definitions (for REST /mcp info)
├── prompts.ts              # Prompt template logic
├── wikipedia-client.ts     # Wikipedia Action API client
├── sanitize.ts             # Input sanitization
├── middleware.ts           # Rate limiting, Server header removal
├── utils.ts                # Shared helpers
└── types.ts                # TypeScript interfaces

Transport design

The server uses stateless Streamable HTTP — the correct mode for serverless platforms:

• POST /mcp — receives JSON-RPC messages, returns JSON or SSE stream
• GET /mcp — opens an SSE channel for server-initiated messages (clients reconnect on timeout)
• DELETE /mcp — session termination (acknowledged; no server-side sessions to clean up)

A fresh McpServer + StreamableHTTPServerTransport instance is created for each request. All Wikipedia logic is handled by the shared, stateless MCPServer helper.

Development

npm install
npm run dev       # tsx watch mode on port 8000
npm run build     # tsc → dist/
npm start         # node dist/server.js
vercel dev        # local Vercel simulation

Bot authentication

For higher Wikipedia API rate limits, create a bot password:

1. Log in to your Wikipedia account
2. Go to Special:BotPasswords
3. Create a new bot password
4. Set WIKIPEDIA_BOT_USERNAME=YourAccount@BotName and WIKIPEDIA_BOT_PASSWORD=...

Quick Start

1. Deploy

git clone https://github.com/YOUR_USERNAME/wikipedia-mcp-vercel.git
cd wikipedia-mcp-vercel
vercel

Or use the Deploy with Vercel button above.

2. Connect your AI client

The server implements MCP Streamable HTTP transport (2025-03-26).
Both the POST channel (client → server) and the GET SSE channel (server → client) are supported.

MCP endpoint: https://your-app.vercel.app/mcp

Claude Desktop

{
  "mcpServers": {
    "wikipedia": {
      "type": "http",
      "url": "https://your-app.vercel.app/mcp"
    }
  }
}

Claude Code / CLI

claude mcp add wikipedia --transport http https://your-app.vercel.app/mcp

Any MCP client (generic)

Configure the client with transport type streamable-http and the endpoint URL above.
The server uses stateless mode (no Mcp-Session-Id header), which is correct and expected for serverless deployments.

3. Environment variables (optional)

Copy .env.example → .env for local development.

4. Verify

# Health check
curl https://your-app.vercel.app/health

# Test MCP initialize
curl -X POST https://your-app.vercel.app/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-03-26","capabilities":{},"clientInfo":{"name":"test","version":"1.0"}}}'

License

MIT