camoufox-mcp-python
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@camoufox-mcp-python Go to https://example.com and take a screenshot"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
camoufox-mcp-python
A Model Context Protocol (MCP) server that provides browser automation tools powered by Camoufox — a stealthy, anti-detect browser built on Firefox. It exposes a Playwright-MCP compatible tool interface using accessibility snapshots and ref-based element targeting.
Think of it as playwright-mcp, but with built-in anti-detection and fingerprint spoofing.
Features
Anti-detect browser fingerprinting — Camoufox automatically spoofs browser fingerprints to bypass bot detection
Ref-based accessibility snapshots — Interact with page elements via accessibility refs, no CSS selectors needed
Human-like cursor movement — Cursor humanization enabled by default for realistic interactions
WebRTC leak protection — WebRTC blocked by default to prevent IP leaks
Isolated browser profiles by default — Each server process gets a fresh temporary profile that is deleted on close; pass
--user-data-dirto reuse a persistent profileGeoIP auto-matching — Automatically match browser locale/timezone to proxy location
Proxy support — Full proxy support including authentication
WebGL fingerprint control — Specify WebGL vendor/renderer pair or block entirely
Custom addons — Load custom Firefox addons or exclude defaults like uBlock Origin
Advanced fingerprint config — Override any BrowserForge fingerprint property or use a custom fingerprint
Firefox preferences — Set custom Firefox user preferences
Browser caching — Optional page/request caching for performance
Privacy controls — Block WebGL and image loading to reduce leaks and bandwidth
Multi-tab management — Create, close, select, and list browser tabs
Modal & dialog handling — Handle JavaScript dialogs and file choosers
Console & network monitoring — Inspect console messages and network requests
Related MCP server: Camoufox MCP
Quick Start
Use with Claude Desktop
Add to your Claude Desktop configuration (claude_desktop_config.json):
{
"mcpServers": {
"camoufox": {
"command": "uvx",
"args": ["camoufox-mcp-python"]
}
}
}Use with Claude Code
claude mcp add camoufox -- uvx camoufox-mcp-pythonUse with VS Code / Cursor
Add to .vscode/mcp.json:
{
"servers": {
"camoufox": {
"command": "uvx",
"args": ["camoufox-mcp-python"]
}
}
}Use with Codex CLI
Add to codex.json or pass via CLI:
codex --mcp-server "npx @anthropic-ai/mcp-proxy -- uvx camoufox-mcp-python"Pin a specific version
{
"mcpServers": {
"camoufox": {
"command": "uvx",
"args": ["--from", "camoufox-mcp-python==0.2.1", "camoufox-mcp-python"]
}
}
}CLI Options
Option | Description |
| Run the browser headlessly (default: on) |
| Proxy server URL (supports |
| Target OS fingerprint: |
| Cursor humanization (default: on). Pass a number for max duration, or |
| Auto-match locale/timezone to proxy IP, or specify a target IP |
| Browser locale(s), e.g. |
| Outer window size, e.g. |
| Block WebRTC to prevent IP leaks (default: on) |
| Disable WebGL |
| Block image requests to reduce bandwidth |
| Disable COOP for cross-origin iframe interactions |
| Optional persistent profile directory. If omitted, each server process uses an isolated temporary profile that is deleted on close |
| Enable capability groups, e.g. |
| WebGL vendor/renderer pair, e.g. |
| Paths to extracted Firefox addons (repeatable, comma-separated) |
| Default addons to exclude, e.g. |
| Camoufox fingerprint properties as a JSON string |
| Cache previous pages and requests (uses more memory) |
| Firefox user preferences as a JSON string |
| Silence Camoufox warnings for advanced configurations |
| Print the config being sent to Camoufox |
Examples
# Zero-config anti-detection (headless + humanize + block-webrtc on by default)
camoufox-mcp-python
# With proxy and GeoIP matching
camoufox-mcp-python --proxy http://user:pass@proxy.example.com:8080 --geoip
# Visible browser for debugging
camoufox-mcp-python --no-headless --debug
# Custom WebGL fingerprint
camoufox-mcp-python --webgl-config "Intel Inc.,Intel(R) UHD Graphics 620" --os windows
# Custom fingerprint properties
camoufox-mcp-python --config '{"navigator.hardwareConcurrency": 8}'
# With caching and custom Firefox prefs
camoufox-mcp-python --enable-cache --firefox-user-prefs '{"dom.webnotifications.enabled": false}'
# Privacy-hardened mode
camoufox-mcp-python --block-webgl --block-images
# Disable default anti-detection for debugging
camoufox-mcp-python --no-headless --humanize false --no-block-webrtc --debug
# Enable JavaScript evaluation
camoufox-mcp-python --caps dangerousAvailable Tools
Navigation
Tool | Description |
| Navigate to a URL |
| Go back to the previous page |
Snapshot & Interaction
Tool | Description |
| Capture an accessibility snapshot of the current page |
| Click an element by ref (supports double-click, right-click, modifiers) |
| Hover over an element |
| Drag from one element to another |
| Select option(s) in a dropdown |
Input
Tool | Description |
| Type text into an editable element (supports slow typing and submit) |
| Press a keyboard key or key combination |
| Fill multiple form fields in a single call |
Page
Tool | Description |
| Take a screenshot (viewport, full page, or element) |
| Retrieve collected console messages |
| List network requests seen by the page |
| Resize the viewport |
| Wait for text to appear/disappear, or a fixed duration |
| Close the browser session |
Tabs
Tool | Description |
| List, create, close, or select browser tabs |
Modals & Files
Tool | Description |
| Accept or dismiss a JavaScript dialog |
| Handle a file chooser and upload files |
Dangerous (requires --caps dangerous)
Tool | Description |
| Execute arbitrary JavaScript on the page or an element |
How It Works
The server uses Camoufox (a Firefox-based anti-detect browser) through Playwright's async API. Pages are represented as accessibility snapshots in YAML format, where each interactive element is assigned a ref identifier. The AI model reads the snapshot, identifies the target ref, and calls the appropriate tool — no fragile CSS/XPath selectors required.
AI Model ←→ MCP Server ←→ Camoufox Browser
│
├── Accessibility Snapshot (YAML)
├── Ref-based element targeting
└── Anti-detect fingerprintingLocal Development
# Clone and install
git clone https://github.com/user/camoufox-mcp.git
cd camoufox-mcp
uv pip install --python .venv/bin/python -e .
# Run the server
./.venv/bin/camoufox-mcp-python --headless
# Or run as a module
./.venv/bin/python -m camoufox_mcp --headless
# Smoke test
./.venv/bin/python -m compileall camoufox_mcp
./.venv/bin/camoufox-mcp-python --helpRequirements
Python >= 3.10
camoufox >= 0.4.11
mcp[cli] >= 1.25
playwright >= 1.58
Note: The first launch on a new machine may trigger automatic Camoufox binary and addon downloads.
License
Available Tools
19 toolsbrowser_clickC
Perform click on a web page.
Inputs: ref (required), button, element, modifiers, +1 more.
browser_closeA
Close the current browser session.
Inputs: no named parameters.
browser_console_messagesC
Return collected console messages.
Inputs: level, filename.
browser_dragD
Drag from one ref to another.
Inputs: endRef (required), startRef (required), endElement (required), startElement (required).
browser_file_uploadD
Handle an active file chooser.
Inputs: paths.
browser_fill_formB
Fill multiple form fields in one call.
Inputs: fields (required).
browser_handle_dialogC
Accept or dismiss an active dialog.
Inputs: accept (required), promptText.
browser_hoverC
Hover over a web page element.
Inputs: ref (required), element.
browser_network_requestsB
List network requests seen by the page.
Inputs: filename, includeStatic.
browser_press_keyC
Press a key on the keyboard.
Inputs: key (required).
browser_resizeB
Resize the current page viewport.
Inputs: width (required), height (required).
browser_select_optionC
Select one or more values in a dropdown.
Inputs: ref (required), values (required), element.
browser_snapshotC
Capture the current accessibility snapshot.
Inputs: filename.
browser_tabsC
List, create, close, or select a browser tab.
Inputs: action (required), index.
browser_take_screenshotB
Take a screenshot of the viewport, full page, or a referenced element.
Inputs: ref, type, element, filename, +1 more.
browser_typeC
Type text into an editable element.
Inputs: ref (required), text (required), slowly, submit, +1 more.
browser_wait_forB
Wait for text to appear or disappear, or for a fixed duration.
Inputs: text, time, textGone.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 19 tools
Each tool targets a distinct interaction or observation (e.g., click vs. hover vs. drag, type vs. press_key vs. fill_form). The only mild overlap is between type and fill_form, but their descriptions clearly scope them to single-element vs. multi-field actions. Overall, no two tools are easily confused.
All tools share the 'browser_' prefix, creating a strong family resemblance. However, the suffix pattern is mixed: some are verb-first (browser_click, browser_navigate) while others are resource-first (browser_tabs, browser_console_messages, browser_network_requests). Despite this, the prefix and clear action descriptions make the naming predictable.
At 19 tools, the set is slightly above the typical 3-15 range but well-justified for a comprehensive browser automation server. The count covers navigation, input, observation, and tab/window management without feeling redundant. A few tools could theoretically be combined (e.g., screenshot modes), but the granularity is reasonable.
The tool surface covers the core browser automation lifecycle: navigation (navigate, back), interaction (click, hover, drag, select, type, press_key, fill_form), waiting (wait_for), and introspection (snapshot, console, network, screenshot). It also handles dialogs, file uploads, tabs, resize, and closing, leaving no obvious dead ends for standard automation scenarios. Advanced features like JS evaluation are absent but not essential for basic browser control.
Related MCP Connectors
Official Browserless MCP server: browser automation, scraping, screenshots and PDFs for AI agents.
Web scraping MCP server — scrape, extract structured data, screenshot any site with anti-bot bypass.
MCP server for Mint — AI-powered QA that runs your app in a real browser on every PR.
Web scraping, browser automation, screenshots and AI data extraction for AI agents via MCP.
Related MCP Servers
- AlicenseBqualityNot gradedmaintenanceA MCP server that provides browser automation tools, allowing users to navigate websites, take screenshots, click elements, fill forms, and execute JavaScript through Playwright.82Apache 2.0
- AlicenseBqualityDmaintenanceAn MCP server for anti-detection browser automation that uses Camoufox to bypass bot detection and spoof digital fingerprints. It enables AI agents to perform human-like web interactions, including realistic cursor movements, humanized click delays, and automatic cookie popup dismissal.36110 npm7MIT
- AlicenseNot gradedqualityDmaintenanceMCP server for controlling a local camofox-browser instance, enabling LLM agents to perform web automation tasks such as navigation, interaction, snapshotting, and content extraction.92 npmMIT
- AlicenseBqualityDmaintenanceBrowser automation MCP server using Camoufox anti-detect browser with fingerprint spoofing, geolocation/timezone spoofing, and human-like cursor movement.141Apache 2.0