inetworkx/web-reader-mcp

このページは AI により自動翻訳されています。技術的な詳細は原文をご確認ください。原文を見る

3 stars 7 forks 品質 62 トレンド 10

MCP server that fetches URLs and extracts clean, readable content as Markdown or plain text. Built on the Model Context Protocol so AI assistants (Claude, Cline, OpenCode, etc.) can read web pages.

概要

MCP server that fetches URLs and extracts clean, readable content as Markdown or plain text. Built on the Model Context Protocol so AI assistants (Claude, Cline, OpenCode, etc.) can read web pages. - Extracts main article content from any URL using Readability - Converts HTML to clean Markdown via Turndown - Extracts metadata (title, author, publish date, Open Graph tags) - Optionally extracts all links from the page - Optional JavaScript rendering via Playwright for SPAs and JS-heavy sites - Exposes a single webReader tool over MCP Streamable HTTP transport The server starts on http://localhost:3001 (or PORT env var). By default, pages are fetched with a simple HTTP client (got). For JavaScript-heavy sites (SPAs, React apps), enable Playwright: Then set render_js: true when calling the tool. Note that Playwright requires ~300–500 MB RAM per browser tab. Runs on port 3000 with health checks enabled. - : Node.

README

以下は原文 README の抜粋です。原文を見る。

web-reader-mcp

MCP server that fetches URLs and extracts clean, readable content as Markdown or plain text. Built on the Model Context Protocol so AI assistants (Claude, Cline, OpenCode, etc.) can read web pages.

Features

Extracts main article content from any URL using Readability
Converts HTML to clean Markdown via Turndown
Extracts metadata (title, author, publish date, Open Graph tags)
Optionally extracts all links from the page
Optional JavaScript rendering via Playwright for SPAs and JS-heavy sites
Exposes a single webReader tool over MCP Streamable HTTP transport

Quick Start

bun install
bun run dev

The server starts on http://localhost:3001 (or PORT env var).

Tool: `webReader`

Parameter	Type	Default	Description
`url`	`string` (required)	—	URL to fetch
`return_format`	`"markdown"` \| `"text"`	`"markdown"`	Output format
`timeout`	`number` (1–60)	`20`	Request timeout in seconds
`retain_images`	`boolean`	`true`	Keep image references in output
`with_links_summary`	`boolean`	`false`	Include a summary of all links
`render_js`	`boolean`	`false`	Use Playwright to render JavaScript

Example Response

{
  "title": "Article Title",
  "url": "https://example.com/article",
  "content": "# Article Title\n\nArticle content in markdown...",
  "excerpt": "A brief summary of the article.",
  "byline": "Author Name",
  "siteName": "Example",
  "publishedTime": "2025-01-15T10:00:00Z",
  "metadata": { "og:description": "..." }
}

MCP Client Configuration

Claude Desktop

{
  "mcpServers": {
    "web-reader": {
      "url": "http://localhost:3001/mcp"
    }
  }
}

Cline (VS Code)

{
  "mcpServers": {
    "web-reader": {
      "url": "http://localhost:3001/mcp",
      "transportType": "streamable-http"
    }
  }
}

JavaScript Rendering (Optional)

By default, pages are fetched with a simple HTTP client (got). For JavaScript-heavy sites (SPAs, React apps), enable Playwright:

bun add playwright
npx playwright install chromium

Then set render_js: true when calling the tool. Note that Playwright requires ~300–500 MB RAM per browser tab.

Scripts

Command	Description
`bun run dev`	Development server with hot reload
`bun run build`	Compile TypeScript
`bun run start`	Production server (`node dist/index.js`)
`bun run typecheck`	Type-check without emitting

Docker

docker compose up --build

Runs on port 3000 with health checks enabled.

Architecture

src/
├── index.ts              # Express server + MCP session management
├── types.ts              # Shared TypeScript interfaces
├── tools/

GitHub で README 全文を見る

インストール

npx playwright install chromium

MCP 初心者？mcpskill.org のガイドから始めましょう →

設定

{
  "mcpServers": {
    "web-reader": {
      "url": "http://localhost:3001/mcp"
    }
  }
}