MCP Server for Web Crawling

An MCP (Model Context Protocol) server that provides web crawling capabilities using crawl4ai. Supports multiple content formats output (HTML, JSON, PDF, screenshots, Markdown) and browser interaction features.

Quick Start

To launch the server, you have several options:

Option 1: Using the Python launch script

Option 2: Using the shell script (Unix-based systems)

Option 3: Using the batch script (Windows)

Option 4: Direct execution

Table of Contents

• MCP Configuration

• Features

• Architecture

• Installation

• Configuration

• Usage

• Testing

• Deployment

• Troubleshooting

• License

MCP Configuration

To configure this server with MCP, you have multiple options depending on your setup:

Method 1: Using the launch script (Recommended)

This is the recommended approach as it handles virtual environment setup and dependency installation automatically:

Make sure to replace /absolute/path/to/your/dev-tool-mcp with the actual absolute path to your project directory.

Method 2: Using the virtual environment Python directly

If you prefer to run the server directly with the virtual environment:

Method 3: Using the installed console script

If the package is installed in the virtual environment, you can use the console script:

Method 4: Windows Configuration

For Windows systems, use the batch script:

Note: The launch scripts (Methods 1-3 on Unix/Linux/Mac and Method 4 on Windows) are recommended because they will:

• Check if the virtual environment exists

• Create it if needed

• Install dependencies from pyproject.toml

• Activate the environment

• Start the server with all necessary dependencies

This ensures a consistent and reliable setup for the MCP server.

Features

• Web Crawling: Advanced web crawling capabilities using crawl4ai

• Multiple Output Formats: Support for HTML, JSON, PDF, screenshots, and Markdown output

• Browser Interaction: Get page content, console messages, and network requests

• LLM Integration: Supports LLM extraction strategies for content processing

• File Download: Automatic download and saving of files found on crawled pages

• Progress Tracking: Streaming progress updates during crawling operations

• Security: URL validation and sanitization to prevent security issues

Architecture

The MCP server is structured into the following modules:

Core Components

• Server: Implements MCP protocol with tool registration and execution

• BrowserService: Manages Playwright browser instances for page content and network monitoring

• Crawler: Uses crawl4ai for advanced web crawling with multiple output formats

• Utils: Provides file handling utilities for saving content

Installation

Prerequisites

• Python 3.8 or higher

• pip package manager

• System dependencies for Playwright (Chromium browser)

Steps

1. Clone the repository:

   git clone <repository-url> cd mcp-server

2. Install the package and dependencies:

   pip install -e .

3. Install Playwright browsers:

   python -m playwright install chromium

Dependencies

The server requires the following dependencies (automatically installed via pip):

• crawl4ai>=0.7.7: Web crawling library

• pydantic>=2.0.0: Data validation

• mcp==1.0.0: Model Context Protocol implementation

• httpx[socks]: HTTP client

• litellm: LLM interface

• beautifulsoup4>=4.12.2: HTML parsing

• lxml>=4.9.3: XML/HTML processing

• sentencepiece: Text processing

• playwright>=1.40.0: Browser automation

Configuration

Environment Variables

The server uses the following environment variables (optional):

• TEST_URL: URL for testing (used in test files)

Available Tools

The server exposes the following tools via MCP:

say_hello

• Description: A simple greeting tool that returns personalized messages to users

• Parameters:

  • name (string, optional): The name to greet, defaults to "World"

• Returns: Greeting message text

echo_message

• Description: Echo tool that returns user-provided information as-is

• Parameters:

  • message (string, required): The message to echo back

• Returns: Echoed message text

crawl_web_page

• Description: Crawl web page content and save in multiple formats (HTML, JSON, PDF, screenshots) while downloading file resources from the page

• Parameters:

  • url (string, required): The URL of the web page to crawl

  • save_path (string, required): The base file path to save the crawled content and downloaded files

  • instruction (string, optional): The instruction to use for the LLM (default: "")

  • save_screenshot (boolean, optional): Save a screenshot of the page (default: false)

  • save_pdf (boolean, optional): Save a PDF of the page (default: false)

  • generate_markdown (boolean, optional): Generate a Markdown representation of the page (default: false)

• Returns: Success message with file count and save location

get_page_content

• Description: Get complete content of a specified URL webpage, including HTML structure and page data

• Parameters:

  • url (string, required): The URL of the web page to get content from

  • wait_for_selector (string, optional): Optional CSS selector to wait for before getting content

  • wait_timeout (integer, optional): Wait timeout in milliseconds, default 30000

• Returns: JSON object containing page content, title, HTML, text, metadata, links, and images

get_console_messages

• Description: Capture console output information from specified URL webpage (including logs, warnings, errors, etc.)

• Parameters:

  • url (string, required): The URL of the web page to get console messages from

  • wait_for_selector (string, optional): Optional CSS selector to wait for before getting console messages

  • wait_timeout (integer, optional): Wait timeout in milliseconds, default 30000

• Returns: JSON object containing console messages with type, text, location, and stack information

get_network_requests

• Description: Monitor and retrieve all network requests initiated by specified URL webpage (API calls, resource loading, etc.)

• Parameters:

  • url (string, required): The URL of the web page to get network requests from

  • wait_for_selector (string, optional): Optional CSS selector to wait for before getting network requests

  • wait_timeout (integer, optional): Wait timeout in milliseconds, default 30000

• Returns: JSON object containing requests and responses with URLs, status, headers, and timing information

Usage

Running the Server

The server can be started using the console script defined in pyproject.toml:

Or directly via Python:

The server uses stdio for MCP communication, making it compatible with MCP clients.

Tool Examples

Crawling a Web Page

To crawl a web page and save content in multiple formats:

This will create a timestamped subdirectory with:

• output.html - Page HTML content

• output.json - Page content in JSON format

• output.png - Screenshot of the page (if requested)

• output.pdf - PDF of the page (if requested)

• raw_markdown.md - Markdown representation of the page (if requested)

• downloaded_files.json - List of downloaded files

• files/ - Directory containing downloaded files

Getting Page Content

To retrieve page content:

Getting Console Messages

To capture console messages from a page:

Getting Network Requests

To monitor network requests made by a page:

Testing

The project includes comprehensive tests for both browser and crawler functionality:

Run Tests

Test Coverage

• test_crawler.py: Tests the complete crawler functionality, including file saving and format generation

• test_browser.py: Tests browser service functions for page content, console messages, and network requests

The crawler test specifically verifies:

• HTML, JSON, PDF, screenshot, and Markdown file generation

• File download and saving functionality

• Proper error handling and directory creation

Deployment

Production Deployment

1. Install the package in your production environment:

   pip install dev-tool-mcp

2. Install Playwright browsers:

   python -m playwright install chromium --with-deps

3. Run the server:

   dev-tool-mcp

Docker Deployment (Optional)

Create a Dockerfile for containerized deployment:

Security

URL Validation

The server includes security measures to prevent malicious URLs:

• URL length limit (2048 characters)

• Protocol validation (only http/https allowed)

• Input sanitization to prevent injection attacks

• Validation to prevent access to local network addresses

Browser Security

• Playwright runs in headless mode by default

• Security flags are set to disable dangerous features

• Browser runs with limited privileges

File System Security

• Files are saved only to explicitly specified paths

• No arbitrary file system access is allowed

• Temporary files are properly cleaned up

Troubleshooting

Common Issues

Playwright Browser Not Found

If you encounter browser-related errors:

Permission Errors

Ensure the server has write permissions to the specified save paths:

Network Issues

For network-related crawling issues, verify:

• The target URL is accessible

• No firewall restrictions exist

• Appropriate timeout values are set

Memory Usage

Large page crawls can consume significant memory. For large-scale crawling:

• Monitor memory usage

• Process pages sequentially rather than in parallel

• Clean up temporary files regularly

Debugging

Enable verbose logging to troubleshoot issues:

Version Information

• Package: dev-tool-mcp

• Version: 0.1.0

• Python Support: 3.8+

• Dependencies: See pyproject.toml for exact versions

License

This project is licensed under the terms specified in the pyproject.toml file. See the project repository for full license details.