PyPen MCP
# PyPen MCP
Model Context Protocol server for browser-based penetration testing using PyDoll.
## Overview
PyPen MCP provides a comprehensive set of tools for LLM-powered browser automation and security testing. It leverages PyDoll's Chrome DevTools Protocol integration for reliable, webdriver-free browser control.
## Features
### Browser Management
- Launch/close browser instances with configurable options
- Navigation control (navigate, back, forward, refresh)
- Headless mode, proxy support, custom user agents
- Incognito mode and resource blocking
### Network Operations
- **Monitoring**: Capture all HTTP traffic, analyze requests/responses
- **Interception**: Block, modify, or mock requests in real-time
- **Authenticated Requests**: Make API calls that inherit browser session/cookies
### DOM Manipulation
- Find elements by CSS selector, XPath, or attributes
- Click, type, scroll, and interact with elements
- Extract page source, element text, and HTML
- Human-like scrolling with physics simulation
### JavaScript Execution
- Execute arbitrary JavaScript in browser context
- Access localStorage and sessionStorage
- Extract global variables, forms, and links
### Session Management
- Cookie manipulation (get, set, delete, clear)
- Import/export cookies in multiple formats
- Session-aware HTTP requests
### Captcha Handling
- **Turnstile Bypass**: Automatic Cloudflare Turnstile interaction
- **Human Intervention**: Request human help for unsolvable captchas (ReCAPTCHA v2, hCaptcha)
- **Detection**: Identify captcha types on the page
### Debugging
- Screenshots (full page or elements)
- Viewport control
- Performance metrics
- Element highlighting
## Installation
```bash
pip install pypen-mcp
```
Or install from source:
```bash
git clone https://github.com/your-repo/pypen-mcp
cd pypen-mcp
pip install -e .
```
## Configuration
Add to your MCP client configuration:
```json
{
"mcpServers": {
"pypen": {
"command": "pypen-mcp"
}
}
}
```
For development:
```json
{
"mcpServers": {
"pypen": {
"command": "/path/to/pypen-mcp/.venv/bin/pypen-mcp"
}
}
}
```
## Usage Examples
### Basic Browser Automation
```python
# Launch browser
browser_launch(headless=True)
# Navigate to target
browser_navigate(url="https://example.com")
# Extract page content
dom_get_source()
# Find and interact with elements
dom_find_element(css_selector="#username")
dom_type(selector="#username", text="admin")
dom_click(selector="button[type=submit]")
# Take screenshot for documentation
debug_screenshot(path="/tmp/screenshot.png")
# Close when done
browser_close()
```
### Network Monitoring
```python
# Enable traffic capture
network_enable_monitoring()
# Navigate and capture
browser_navigate(url="https://api.example.com")
# Get captured requests
network_get_logs(filter_url="/api/")
# Get response body for specific request
network_get_response_body(request_id="abc123")
# Disable when done
network_disable_monitoring()
```
### Request Interception
```python
# Set up interception rules
network_setup_handler(
block_patterns=["analytics", "tracking", "ads"],
modify_headers={"Authorization": "Bearer token123"},
mock_responses={"/api/config": {"status": 200, "body": {"debug": true}}}
)
# Navigate with interception active
browser_navigate(url="https://example.com")
```
### Session-Based Requests
```python
# Login via UI (handles complex auth)
browser_navigate(url="https://app.com/login")
dom_type(selector="#email", text="user@example.com")
dom_type(selector="#password", text="password123")
dom_click(selector="button[type=submit]")
# Make authenticated API calls
session_make_request(
url="https://app.com/api/user/profile",
method="GET"
)
```
### Captcha Handling
```python
# Enable automatic Turnstile bypass
captcha_enable_turnstile_bypass(time_to_wait_captcha=10.0)
# Navigate to protected site
browser_navigate(url="https://protected-site.com")
# Wait for captcha processing
# ... automation continues ...
# For unsolvable captchas, request human help
captcha_detect_type() # Check what's on page
captcha_request_human_intervention(
captcha_type="recaptcha_v2",
message="Please solve the ReCAPTCHA to continue"
)
# Wait for human to solve
captcha_wait_for_resolution(request_id="abc123", timeout=300)
```
## Tool Reference
### Browser Tools
| Tool | Description |
|------|-------------|
| `browser_launch` | Launch browser with options (headless, proxy, etc.) |
| `browser_close` | Close the browser instance |
| `browser_navigate` | Navigate to a URL |
| `browser_go_back` | Navigate back in history |
| `browser_go_forward` | Navigate forward in history |
| `browser_refresh` | Refresh current page |
| `browser_get_info` | Get current URL and title |
### Network Tools
| Tool | Description |
|------|-------------|
| `network_enable_monitoring` | Start capturing HTTP traffic |
| `network_disable_monitoring` | Stop traffic capture |
| `network_get_logs` | Get captured requests (with optional filter) |
| `network_get_response_body` | Get response body for a request |
| `network_enable_interception` | Enable request interception |
| `network_disable_interception` | Disable interception |
| `network_setup_handler` | Configure auto interception rules |
### DOM Tools
| Tool | Description |
|------|-------------|
| `dom_find_element` | Find element by selector |
| `dom_find_elements` | Find multiple elements |
| `dom_get_text` | Get element text content |
| `dom_get_html` | Get element HTML |
| `dom_click` | Click an element |
| `dom_type` | Type text into input |
| `dom_scroll` | Scroll the page |
| `dom_get_source` | Get full page HTML |
| `dom_wait_for` | Wait for element to appear |
### JavaScript Tools
| Tool | Description |
|------|-------------|
| `js_execute` | Execute JavaScript code |
| `js_get_console_logs` | Get console logs |
| `js_get_global_vars` | Extract global variables |
| `js_get_local_storage` | Get localStorage data |
| `js_get_session_storage` | Get sessionStorage data |
| `js_set_local_storage` | Set localStorage value |
| `js_get_forms` | Get all forms with inputs |
| `js_get_links` | Get all links on page |
### Session Tools
| Tool | Description |
|------|-------------|
| `session_get_cookies` | Get all cookies |
| `session_get_cookie` | Get specific cookie |
| `session_set_cookie` | Set a cookie |
| `session_delete_cookie` | Delete a cookie |
| `session_clear_cookies` | Clear all cookies |
| `session_export_cookies` | Export cookies (json/netscape/header) |
| `session_import_cookies` | Import cookies from JSON |
| `session_make_request` | Make session-authenticated HTTP request |
### Captcha Tools
| Tool | Description |
|------|-------------|
| `captcha_enable_turnstile_bypass` | Enable auto Turnstile handling |
| `captcha_disable_turnstile_bypass` | Disable Turnstile handling |
| `captcha_detect_type` | Detect captcha type on page |
| `captcha_handle_auto` | Auto-handle detected captcha |
| `captcha_request_human_intervention` | Request human help |
| `captcha_get_pending_interventions` | Get unresolved requests |
| `captcha_resolve_intervention` | Mark intervention as resolved |
| `captcha_wait_for_resolution` | Wait for human to solve |
### Debug Tools
| Tool | Description |
|------|-------------|
| `debug_screenshot` | Take page screenshot |
| `debug_get_viewport` | Get viewport size |
| `debug_set_viewport` | Set viewport size |
| `debug_get_performance` | Get performance metrics |
| `debug_highlight` | Highlight matching elements |
## Important Notes
### Browser Session Persistence
- The browser session persists until explicitly closed with `browser_close`
- All operations (DOM, JS, network) use the same active session
- Cookies and session state are maintained across navigations
### Captcha Success Factors
Turnstile bypass success depends on:
1. **IP Reputation**: Use residential proxies with good reputation
2. **Browser Fingerprint**: Configure realistic browser options
3. **Behavioral Patterns**: Add realistic delays and interactions
### Human Intervention Workflow
1. Detect captcha: `captcha_detect_type()`
2. Try auto-handling: `captcha_handle_auto()`
3. If fails, request help: `captcha_request_human_intervention()`
4. Wait for resolution: `captcha_wait_for_resolution()`
5. Continue automation
## License
MIT
TDQS
Scored across 52 tools
Each tool has a clear, distinct purpose, with prefixes like browser_, captcha_, debug_, dom_, js_, network_, session_ grouping related functionality. No two tools perform the same action, and even within groups, tools are well-differentiated.
All tool names follow a consistent snake_case convention with a prefix indicating the functional domain. This makes it easy for an agent to predict tool names and understand their roles.
52 tools is on the high side, but the server covers a broad scope of browser automation features (browser control, DOM, JS, network, sessions, captchas). The count is appropriate for the comprehensive capability set, though slightly above the ideal range.
The tool surface is remarkably complete: browser lifecycle, navigation, DOM interactions, debugging, JavaScript execution, network monitoring/interception, cookie/session management, and captcha handling all have dedicated tools. No obvious gaps for the intended domain.