wikipedia-mcp-server

wikipedia-mcp-server

0
cloud_platforms

This project is a Cloudflare Worker that provides a Model Context Protocol (MCP) compliant interface to access Wikipedia search functionalities. By leveraging JSON-RPC 2.0, it offers structured methods for searching articles and retrieving page content by title or ID. Its deployment is focused on simplicity and efficiency for the user.

Wikipedia MCP Server (Cloudflare Worker)

A Cloudflare Worker exposing Wikipedia search and page retrieval via the Model Context Protocol (MCP) standard using JSON-RPC 2.0.

Features

  • MCP Endpoint: POST /mcp endpoint supporting JSON-RPC 2.0 requests for Wikipedia operations.
    • wikipedia.search: Search for articles.
    • wikipedia.page: Get page content by title.
    • wikipedia.pageById: Get page content by ID.
  • Configurable LRU Cache (In-memory per instance, consider KV/Caches API for persistence).
  • Written in TypeScript with Hono.
  • Deployable to Cloudflare Workers.

Requirements

  • Node.js (check .nvmrc or package.json engines if specified)
  • npm or yarn
  • Wrangler CLI (Cloudflare's Worker CLI)
  • A Cloudflare account

Installation

git clone https://github.com/your-repo/wikipedia-mcp-server.git # Replace with your repo URL
cd wikipedia-mcp-server
npm install
npm install -g wrangler # Or install locally: npm install --save-dev wrangler
wrangler login # Authenticate with Cloudflare

Configuration

Worker configuration is managed via wrangler.toml.

  • name: Name of the worker on Cloudflare.
  • main: Entry point script (dist/worker.js after build).
  • compatibility_date: Sets the runtime compatibility date.
  • [build]: Specifies the build command (npm run build).

Local Development Variables:

Create a .dev.vars file in the wikipedia directory root for local development using wrangler dev. Wrangler automatically loads this file. Example:

# .dev.vars (optional, for local development overrides)
CACHE_MAX=200
CACHE_TTL=600000 # 10 minutes

Deployed Variables (Secrets & Environment Variables):

For deployed workers, configure environment variables and secrets via the Cloudflare dashboard or using wrangler secret put VAR_NAME. These might be needed if extending functionality (e.g., API keys for other services).

Running Locally

npm run dev

This starts a local server using Wrangler, simulating the Cloudflare environment. It typically listens on http://localhost:8787. Check the Wrangler output for the exact address. The /mcp endpoint will be available at this address.

Deployment

npm run deploy

This command builds the worker using npm run build (as defined in wrangler.toml) and deploys it to your Cloudflare account. Wrangler will output the URL of your deployed worker.

API Endpoint

POST /mcp (JSON-RPC 2.0)

The single endpoint /mcp accepts POST requests with a JSON-RPC 2.0 payload.

Request Format

{
  "jsonrpc": "2.0",
  "method": "<method_name>",
  "params": { ... }, // Method-specific parameters
  "id": "<request_id>"
}

Supported Methods

  • wikipedia.search
    • Params: { "query": string, "limit"?: number, "lang"?: string } (Note: offset might need re-implementation if required)
    • Result: Array of search result objects ({ title, snippet, pageid })
  • wikipedia.page
    • Params: { "title": string, "lang"?: string }
    • Result: Page object ({ title, pageid, text, sections }) or null
  • wikipedia.pageById
    • Params: { "id": number, "lang"?: string }
    • Result: Page object ({ title, pageid, text, sections }) or null

Response Format (Success/Error): Standard JSON-RPC 2.0 success/error objects.

Integrations

(This section remains largely the same, just update example URLs if needed)

This MCP-compliant Wikipedia server can be integrated into various tools and platforms that support consuming JSON-RPC 2.0 services over HTTP. The primary integration point is the /mcp endpoint.

Cursor

To integrate this Wikipedia service with Cursor, configure Cursor to make requests to this worker's /mcp endpoint (e.g., https://your-worker-name.your-subdomain.workers.dev/mcp after deployment, or http://localhost:8787/mcp during local development).

The request body should follow the JSON-RPC 2.0 specification:

{
  "jsonrpc": "2.0",
  "method": "wikipedia.search",
  "params": { "query": "Cloudflare Workers", "limit": 5 },
  "id": "cursor-req-123"
}

Refer to Cursor's documentation for specific instructions.

Windsurf

If Windsurf can consume standard JSON-RPC 2.0 endpoints over HTTP, configure it to point to this worker's /mcp endpoint (e.g., https://your-worker-name.your-subdomain.workers.dev/mcp).

Consult the Windsurf documentation for details.

Testing

Note: The existing tests (src/__tests__/server.test.ts) were designed for the previous Express server and are not compatible with the Cloudflare Worker setup. They need to be rewritten using tools suitable for testing Hono applications within a Worker context, such as:

  • Hono's built-in testing utilities (app.request(...)).
  • vitest with miniflare or a similar Worker environment simulator.

This section needs updates once tests are implemented.

Usage Examples

(Removed REST/GraphQL examples. Updated JSON-RPC examples to use a placeholder worker URL).

JSON-RPC - cURL

(Replace http://localhost:8787/mcp with your deployed worker URL if testing against deployment)

# Search
curl -X POST http://localhost:8787/mcp \\
  -H "Content-Type: application/json" \\
  -d '{
    "jsonrpc": "2.0", 
    "method": "wikipedia.search", 
    "params": { "query": "Hono", "limit": 2 }, 
    "id": "search-1"
  }'

# Get Page by Title
curl -X POST http://localhost:8787/mcp \\
  -H "Content-Type: application/json" \\
  -d '{
    "jsonrpc": "2.0", 
    "method": "wikipedia.page", 
    "params": { "title": "Cloudflare", "lang": "en" }, 
    "id": "page-1"
  }'

# Get Page by ID
curl -X POST http://localhost:8787/mcp \\
  -H "Content-Type: application/json" \\
  -d '{
    "jsonrpc": "2.0", 
    "method": "wikipedia.pageById", 
    "params": { "id": 21721040, "lang": "en" }, 
    "id": "pageid-1"
  }'

JSON-RPC - Node.js (using node-fetch)

(Replace WORKER_URL with your local or deployed worker URL)

import fetch from 'node-fetch';

const WORKER_URL = 'http://localhost:8787/mcp'; // Or your deployed URL

async function jsonRpcRequest(method: string, params: Record<string, any>) {
  const response = await fetch(WORKER_URL, {
    method: 'POST',
    headers: { 'Content-Type': 'application/json' },
    body: JSON.stringify({
      jsonrpc: '2.0',
      method,
      params,
      id: Date.now().toString() // Example ID
    })
  });
  if (!response.ok) {
    throw new Error(`HTTP error! status: ${response.status}`);
  }
  return response.json();
}

async function main() {
  try {
    const searchResults = await jsonRpcRequest('wikipedia.search', { query: 'WebAssembly', limit: 3 });
    console.log('Search:', JSON.stringify(searchResults, null, 2));

    const pageData = await jsonRpcRequest('wikipedia.page', { title: 'TypeScript' });
    console.log('\\nPage:', JSON.stringify(pageData, null, 2));
    
    const pageByIdData = await jsonRpcRequest('wikipedia.pageById', { id: 21721040 }); // Example ID for Node.js page
    console.log('\\nPage By ID:', JSON.stringify(pageByIdData, null, 2));

  } catch (error) {
      console.error('Error making JSON-RPC request:', error);
  }
}

main();