html2md-mcp
Converts HTML webpages to clean Markdown format, reducing size by ~90-95% while preserving tables, images, and important content. Supports both simple HTTP fetch and Playwright browser mode for JavaScript-heavy or authenticated pages.
README
HTML to Markdown MCP Server
MCP (Model Context Protocol) server for converting HTML webpages to clean Markdown format. Reduces HTML size by ~90-95% while preserving tables, images, and important content - perfect for AI context.
Features
- Converts HTML from URLs to clean Markdown
- Preserves tables, images, and links
- Removes unnecessary elements (scripts, styles, navigation, footers, headers)
- Significant size reduction (typically 90-95% compression)
- Configurable options for images, tables, and links
- Built with
trafilaturaandBeautifulSoup4for robust extraction - Stream processing for efficient handling of large pages
- Size limits to prevent downloading excessively large content (1MB-50MB)
- Optional caching to speed up repeated conversions of the same URLs
- 🌐 Browser mode with Playwright - Handles JavaScript-heavy sites and authenticated pages
- Execute JavaScript (perfect for SPAs: React, Vue, Angular)
- Use your browser profile with cookies (access authenticated pages!)
- Support for Chrome, Firefox, WebKit
- Configurable wait strategies for dynamic content
Installation
Prerequisites
- Python 3.10 or higher
uvpackage manager (recommended) orpip
Install with uv (recommended)
# Clone the repository
git clone <your-repo-url>
cd html2md
# Install dependencies
uv pip install -e .
# Install Playwright browsers (required for browser mode)
playwright install chromium
Install with pip
# Clone the repository
git clone <your-repo-url>
cd html2md
# Create virtual environment
python -m venv venv
source venv/bin/activate # On Windows: venv\Scripts\activate
# Install dependencies
pip install -e .
# Install Playwright browsers (required for browser mode)
playwright install chromium
Docker Installation (Recommended for Production)
The easiest way to use html2md is with Docker:
# Build the image
docker build -t html2md .
# Or use pre-built image (when published)
docker pull your-registry/html2md:latest
For Claude Desktop, configure with Docker:
{
"mcpServers": {
"html2md": {
"command": "docker",
"args": [
"run",
"-i",
"--rm",
"html2md"
]
}
}
}
Docker Image Features:
- Pre-installed Playwright with Chromium
- Optimized for minimal size (~1GB)
- Non-root user for security
- Ready to use - no additional setup required
Configuration
Add the server to your Claude Desktop configuration file:
macOS
Edit ~/Library/Application Support/Claude/claude_desktop_config.json:
{
"mcpServers": {
"html2md": {
"command": "uv",
"args": [
"--directory",
"/absolute/path/to/html2md",
"run",
"html2md"
]
}
}
}
Windows
Edit %APPDATA%/Claude/claude_desktop_config.json:
{
"mcpServers": {
"html2md": {
"command": "uv",
"args": [
"--directory",
"C:\\absolute\\path\\to\\html2md",
"run",
"html2md"
]
}
}
}
Linux
Edit ~/.config/Claude/claude_desktop_config.json:
{
"mcpServers": {
"html2md": {
"command": "uv",
"args": [
"--directory",
"/absolute/path/to/html2md",
"run",
"html2md"
]
}
}
}
Usage
Once configured, the MCP server will be available in Claude Desktop. You can use the html_to_markdown tool:
Example 1: Basic conversion
Convert this webpage to markdown: https://example.com/article
Example 2: With options
Use the html_to_markdown tool with:
- url: https://example.com/docs
- include_images: false
- include_tables: true
Example 3: Browser mode for JavaScript-heavy sites
Use the html_to_markdown tool with:
- url: https://spa-application.com
- fetch_method: playwright
- wait_for: networkidle
Example 4: Access authenticated pages
Use the html_to_markdown tool with:
- url: https://private-site.com/dashboard
- fetch_method: playwright
- use_user_profile: true
- browser_type: chromium
Note: For use_user_profile=true, make sure Chrome is closed before running.
Tool Parameters
Basic Parameters:
url(required): URL of the webpage to convertinclude_images(optional, default: true): Include images in Markdowninclude_tables(optional, default: true): Include tables in Markdowninclude_links(optional, default: true): Include links in Markdowntimeout(optional, default: 30): Request timeout in seconds (5-120)
Performance Parameters:
max_size(optional, default: 10MB): Maximum size of content to download in bytes (1MB-50MB)use_cache(optional, default: false): Enable caching for faster repeated conversionscache_ttl(optional, default: 3600): Cache time-to-live in seconds (60-86400)
Browser Mode Parameters:
fetch_method(optional, default: "fetch"): Fetch method - "fetch" (fast) or "playwright" (handles JS, auth)browser_type(optional, default: "chromium"): Browser to use - "chromium", "firefox", or "webkit"headless(optional, default: true): Run browser in headless modewait_for(optional, default: "networkidle"): Wait strategy - "load", "domcontentloaded", or "networkidle"use_user_profile(optional, default: false): Use your browser profile with cookies (requires Chrome closed)
Development
Install development dependencies
uv pip install -e ".[dev]"
Run tests
pytest
Code formatting
# Format with black
black src/ tests/
# Lint with ruff
ruff check src/ tests/
Type checking
mypy src/
Architecture
The project consists of three main modules:
converter.py
Core HTML to Markdown conversion functionality:
fetch_html(): Downloads HTML from URLclean_html(): Removes unnecessary elements with BeautifulSoupconvert_to_markdown(): Converts cleaned HTML to Markdown with trafilaturahtml_to_markdown(): Main workflow combining all steps
server.py
MCP server implementation:
- Registers the
html_to_markdowntool - Handles tool calls and error responses
- Runs async MCP server with stdio transport
utils.py
Utility functions:
- Hash calculation for caching
- Text formatting and truncation
- Domain extraction
- Filename sanitization
cache.py
In-memory caching system:
SimpleCacheclass with TTL support- Global cache instance management
- Automatic expiration of old entries
- Hash-based cache keys for URL + parameters
browser.py
Playwright browser automation:
fetch_html_playwright()- Async browser-based HTML fetching- Support for Chromium, Firefox, WebKit
- User profile integration for authenticated access
- Configurable wait strategies for dynamic content
Troubleshooting
Server not appearing in Claude Desktop
- Check that the path in
claude_desktop_config.jsonis absolute and correct - Restart Claude Desktop completely
- Check Claude Desktop logs for errors
Installation issues
# Verify Python version
python --version # Should be 3.10+
# Try reinstalling dependencies
uv pip install --force-reinstall -e .
Conversion errors
- Timeout errors: Increase the
timeoutparameter - Empty content: Some websites may block automated requests or use JavaScript rendering
- Solution: Use
fetch_method: playwrightto execute JavaScript
- Solution: Use
- Parse errors: The webpage structure may be unusual or malformed
- Content too large: Increase the
max_sizeparameter (up to 50MB) or the page exceeds limits - Cache issues: Disable caching with
use_cache: falseif you need fresh content
Browser mode issues
- Playwright not installed: Run
playwright install chromium - Browser launch fails: Check that you have sufficient permissions and disk space
- User profile error: Make sure Chrome is completely closed before using
use_user_profile: true - Page doesn't load fully: Try different
wait_forstrategies:"load"- fastest, waits for page load event"domcontentloaded"- waits for DOM to be ready"networkidle"- slowest but most reliable, waits for network to be idle
- Authentication not working: Ensure you're using
browser_type: chromiumanduse_user_profile: true
Performance
Typical conversion results:
- Original HTML: ~500KB - 2MB
- Markdown output: ~25KB - 100KB
- Compression: 90-95%
- Processing time: 2-10 seconds (depending on page size and network)
License
MIT
Contributing
Contributions are welcome! Please feel free to submit issues or pull requests.
Credits
Built with:
- MCP SDK - Model Context Protocol
- trafilatura - Web content extraction
- BeautifulSoup4 - HTML parsing
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。