Skip to main content
Glama
Camoufox Playwright MCP

Camoufox Playwright MCP

๐ŸฆŠ Camoufox Playwright MCP

The Playwright MCP, on Camoufox. Same tools, harder to detect.

PyPI version Python Version License Test Suite

The Playwright MCP tool set, in Python, on Camoufox. Built for sites that fingerprint or block the default browser.

Highlights โ€ข Quickstart โ€ข Client Setup โ€ข Connection Modes โ€ข Configuration โ€ข Tool Reference โ€ข Divergences


โšก Highlights

  • ๐ŸฆŠ Camoufox by Default: Runs Camoufox, a Firefox build with fingerprint spoofing built in (navigator.webdriver masked, canvas/WebGL/audio randomized, hardware details spoofed).

  • ๐ŸŽญ Same tool set as @playwright/mcp, runs on Camoufox. Drop-in for the official server's tool contract (browser_navigate, browser_click, browser_snapshot, browser_evaluate, browser_localstorage_*, and the rest). Camoufox masks navigator.webdriver and randomizes canvas, WebGL, and audio per session. Switch to Chromium, Chrome, Firefox, or WebKit when you don't need stealth.

  • ๐Ÿ Python-Native Code Generation: Emits clean Python Playwright code snippets instead of JavaScript strings:

    await page.get_by_role("button", name="Submit").click()
  • ๐Ÿ”„ Dual Engine Flexibility: Run camoufox by default, or switch to standard Playwright engines (chromium, chrome, firefox, webkit), remote CDP endpoints, or browser extensions.

  • ๐Ÿš€ Zero-Install with uvx: Run on-demand in Claude Desktop, Cursor, Windsurf, Cline, OpenCode, or any MCP client without manual virtualenv management.


Related MCP server: wraith-mcp

๐Ÿค” What this is (and isn't)

This server gives you the same tool set as @playwright/mcp: browser_navigate, browser_click, browser_snapshot, and the rest. Drop it into an existing client config and your prompts keep working.

The browser underneath is different. Camoufox is a Firefox build with anti-fingerprinting built in. navigator.webdriver is masked, canvas and WebGL and audio outputs are randomized per session, and hardware details are spoofed. Useful when the site you're driving detects or blocks the default browser. Switch to Chromium, Chrome, Firefox, or WebKit when you don't need that.

Not affiliated with Microsoft or the Playwright team. The browser-level fingerprinting work lives in Camoufox; this server wires it into the MCP tool contract.


๐Ÿš€ Quickstart

Run On-Demand with uvx

No installation required. Needs Python 3.12+ (uvx handles this). The first launch downloads the Camoufox browser (~150MB, one time), so the first tool call can take a few minutes; after that it starts instantly:

# Run the Camoufox browser in headless mode (recommended for AI agents)
uvx camoufox-playwright-mcp --headless

# Pre-download the browser ahead of time ("all" also installs Playwright browsers)
uvx camoufox-playwright-mcp install-browser all

# Run with visible browser window (headed mode)
uvx camoufox-playwright-mcp

# Switch to standard Chrome / Chromium
uvx camoufox-playwright-mcp --browser chrome --headless

# Enable storage & developer capabilities
uvx camoufox-playwright-mcp --headless --caps=storage,devtools,vision

๐Ÿ’ป MCP Client Configuration

1. Claude Desktop

Add this to your claude_desktop_config.json:

  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

  • Windows: %APPDATA%\Claude\claude_desktop_config.json

  • Linux: ~/.config/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "camoufox": {
      "command": "uvx",
      "args": [
        "camoufox-playwright-mcp",
        "--headless"
      ]
    }
  }
}

2. Cursor (~/.cursor/mcp.json)

Add to your Cursor MCP settings (Cursor Settings > MCP > Add New MCP Server):

{
  "mcpServers": {
    "camoufox": {
      "command": "uvx",
      "args": [
        "camoufox-playwright-mcp",
        "--headless"
      ]
    }
  }
}

3. Cline / Roo Code / Windsurf / Zed

{
  "mcpServers": {
    "browser": {
      "command": "uvx",
      "args": [
        "camoufox-playwright-mcp",
        "--headless"
      ]
    }
  }
}

4. OpenCode (opencode.json)

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "camoufox": {
      "type": "local",
      "command": ["uvx", "camoufox-playwright-mcp", "--headless"],
      "enabled": true
    }
  }
}

๐ŸŒ Transports (stdio & Streamable HTTP)

By default, the server uses stdio transport.

Passing --port starts a Streamable HTTP server listening at /mcp:

uvx camoufox-playwright-mcp --headless --host 127.0.0.1 --port 8931

Client configuration for HTTP transport:

{
  "mcpServers": {
    "camoufox": {
      "url": "http://127.0.0.1:8931/mcp"
    }
  }
}

๐Ÿ”Œ Browser Modes & Connections

1. Persistent Profiles (Default)

By default, the server launches a persistent browser context using an auto-created profile under the user cache directory (~/.cache/camoufox-playwright-mcp/), preserving session state, logins, and cookies across runs:

# Explicit persistent profile directory
uvx camoufox-playwright-mcp --user-data-dir ~/.config/my-browser-profile

2. Isolated Ephemeral Sessions

Use --isolated to run with a temporary, in-memory context that leaves no trace on disk:

uvx camoufox-playwright-mcp --isolated --headless

3. Connect to Existing Browser over CDP

Connect to an existing Chromium / Chrome instance started with --remote-debugging-port=9222:

uvx camoufox-playwright-mcp --cdp-endpoint http://localhost:9222

4. Connect to Bound Playwright Endpoint

Connect to a remote Playwright browser server:

uvx camoufox-playwright-mcp --endpoint ws://localhost:3000

5. Playwright Browser Extension Relay

Attach directly to your existing Chrome or Edge browser tabs via the Playwright Browser Extension:

uvx --from 'camoufox-playwright-mcp[extension]' camoufox-playwright-mcp --extension

๐ŸฆŠ Camoufox Configuration

When running with --browser camoufox (default), you can customize Camoufox's browser parameters via a JSON configuration file (--config config.json):

{
  "browser": {
    "provider": "camoufox",
    "camoufoxOptions": {
      "geoip": true,
      "humanize": true,
      "os": "windows",
      "block_images": false,
      "fonts": ["Arial", "Calibri", "Times New Roman"]
    }
  }
}

Key Camoufox Options

Option

Type

Description

geoip

bool

Automatically match timezone, locale, and geolocation to your IP or proxy.

humanize

bool

Add natural, human-like mouse movements and keyboard typing delays.

os

str

Target OS to emulate ("windows", "macos", "linux").

block_images

bool

Block images to optimize network bandwidth and speed up scraping.

webrtc_ip

str

Spoof WebRTC local IP address to prevent real IP leaks.


โš™๏ธ Configuration (JSON & INI)

Configuration is merged with the following precedence (highest priority last):

  1. Built-in defaults (browser: camoufox, action timeout: 5000ms, output: file).

  2. JSON or INI configuration file (via --config or CAMOUFOX_MCP_CONFIG).

  3. Environment variables (CAMOUFOX_MCP_* or PLAYWRIGHT_MCP_*).

  4. Explicit CLI arguments.

INI Configuration Example (camoufox.ini)

capabilities = storage,devtools,vision
console.level = info
timeouts.action = 8000
timeouts.navigation = 45000
browser.contextOptions.viewport = 1280x720

Load with:

uvx camoufox-playwright-mcp --config camoufox.ini

๐ŸŒ Environment Variables

All CLI flags can be set via environment variables:

Variable

Description

CAMOUFOX_MCP_BROWSER

Default browser (camoufox, chrome, chromium, firefox, webkit)

CAMOUFOX_MCP_HEADLESS

Set to true or 1 for headless mode

CAMOUFOX_MCP_ISOLATED

Set to true to use ephemeral isolated contexts

CAMOUFOX_MCP_PROXY_SERVER

Proxy server URL

CAMOUFOX_MCP_CAPS

Comma-separated list of capabilities (storage, devtools, vision, pdf)

CAMOUFOX_MCP_OUTPUT_DIR

Output directory for artifacts (screenshots, downloads)

CAMOUFOX_MCP_CONFIG

Path to JSON/INI configuration file

(Note: PLAYWRIGHT_MCP_* variables are also supported for backward compatibility).


๐Ÿงฐ Available MCP Tools

Same tools as the official Playwright MCP server:

Navigation & Interaction

  • browser_navigate: Navigate to any URL with automatic wait-for-load.

  • browser_click: Click elements using locators, coordinates, or semantic text.

  • browser_type: Fill input fields (fill semantics, Camoufox humanized typing where available).

  • browser_fill_form: Fill multiple form fields in one call.

  • browser_hover, browser_type, browser_press_key: Natural mouse and keyboard interactions.

  • browser_file_upload: Upload files to file input elements.

  • browser_drag, browser_drop: Perform drag-and-drop operations.

Inspection & Output

  • browser_snapshot: Capture full accessibility and semantic tree snapshots.

  • browser_take_screenshot: Capture full-page or element screenshots.

  • browser_evaluate: Safely evaluate JavaScript within the page context.

  • browser_console_messages: Retrieve console logs and error streams.

Storage & State (Enable with --caps storage)

  • browser_localstorage_list, browser_localstorage_get, browser_localstorage_set, browser_localstorage_delete, browser_localstorage_clear

  • browser_sessionstorage_list, browser_sessionstorage_get, browser_sessionstorage_set, browser_sessionstorage_delete, browser_sessionstorage_clear

  • browser_cookies: Get, set, and clear cookies.

  • browser_storage_state, browser_set_storage_state: Save and restore full storage state files.

Tabs & Network

  • browser_tabs: Manage multiple tabs (list, switch, create, close).

  • browser_network: Manage routing and network interception.

  • browser_handle_dialog: Accept or dismiss JavaScript alerts, confirms, and prompts.


๐Ÿ”ฌ Intentional Python Divergences & Limitations

This server is designed to follow the official Playwright MCP public contract while providing a Python-native experience:

  • Python Code Generation: Upstream @playwright/mcp generates JavaScript snippets; this server generates native Python Playwright snippets (await page.get_by_role(...).click()).

  • browser_run_code_unsafe: Executes asynchronous Python Playwright code directly against the active page instance.

  • browser.initPage: Accepts Python-native modules defining init_page(page) or default(page).

  • browser_annotate: Intentionally omitted (upstream relies on a Node.js dashboard daemon).

  • browser_pdf_save: PDF generation is supported when running Headless Chromium engines (--browser chromium --headless).


๐Ÿงช Testing & Conformance

Run local test and quality gates:

# Lint checks
uv run ruff check .

# Type checking (strict mypy across all source and test files)
uv run mypy src tests

# Unit and integration test suite
uv run pytest

Run upstream TypeScript Playwright MCP conformance suite:

cd tests/conformance/upstream
npm ci
npx playwright test --workers=10

๐Ÿ“„ License

Apache License 2.0. See LICENSE for details.

Available Tools

24 tools
browser_clickClickC
Destructive

Perform click on a web page

Inputs: target (required), _meta, button, element, +2 more.

browser_closeClose browserC
Destructive

Close the page

Inputs: _meta.

browser_console_messagesGet console messagesC
Read-only

Returns all console messages

Inputs: all, _meta, level, filename.

browser_dragDrag mouseC
Destructive

Perform drag and drop between two elements

Inputs: endTarget (required), startTarget (required), _meta, endElement, +1 more.

browser_dropDrop files or data onto an elementB
Destructive

Drop files or MIME-typed data onto an element, as if dragged from outside the page. At least one of "paths" or "data" must be provided.

Inputs: target (required), data, _meta, paths, +1 more.

browser_evaluateEvaluate JavaScriptC
Destructive

Evaluate JavaScript expression on page or element

Inputs: function (required), _meta, target, element, +1 more.

browser_file_uploadUpload filesC
Destructive

Upload one or multiple files

Inputs: _meta, paths.

browser_fill_formFill formC
Destructive

Fill multiple form fields

Inputs: fields (required), _meta.

browser_findFind in page snapshotA
Read-only

Search the accessibility snapshot of the current page for text or a regular expression. Returns matching snapshot nodes with a few lines of surrounding context (like search snippets), which is cheaper than capturing the whole snapshot when you only need to locate an element and its ref.

Inputs: text, _meta, regex.

browser_handle_dialogHandle a dialogD
Destructive

Handle a dialog

Inputs: accept (required), _meta, promptText.

browser_hoverHover mouseC
Destructive

Hover over element on page

Inputs: target (required), _meta, element.

browser_navigateNavigate to a URLC
Destructive

Navigate to a URL

Inputs: url (required), _meta.

browser_navigate_backGo backA
Destructive

Go back to the previous page in the history

Inputs: _meta.

browser_network_requestShow network request detailsA
Read-only

Returns full details (headers and body) of a single network request, or a single part if part is set. Use the number from browser_network_requests.

Inputs: index (required), part, _meta, filename.

browser_network_requestsList network requestsA
Read-only

Returns a numbered list of network requests since loading the page. Use browser_network_request with the number to get full details.

Inputs: _meta, filter, static, filename.

browser_press_keyPress a keyC
Destructive

Press a key on the keyboard

Inputs: key (required), _meta.

browser_resizeResize browser windowD
Destructive

Resize the browser window

Inputs: width (required), height (required), _meta.

browser_run_code_unsafeRun Playwright code (unsafe)A
Destructive

Run a Python Playwright code snippet. Unsafe: executes arbitrary Python in the Playwright server process with normal builtins/imports and is RCE-equivalent.

Inputs: code, _meta, filename.

browser_select_optionSelect optionC
Destructive

Select an option in a dropdown

Inputs: target (required), values (required), _meta, element.

browser_snapshotPage snapshotC
Read-only

Capture accessibility snapshot of the current page, this is better than screenshot

Inputs: _meta, boxes, depth, target, +1 more.

browser_tabsManage tabsC
Destructive

List, create, close, or select a browser tab.

Inputs: action (required), url, _meta, index.

browser_take_screenshotTake a screenshotA
Read-only

Take a screenshot of the current page. You can't perform actions based on the screenshot, use browser_snapshot for actions.

Inputs: type, _meta, scale, target, +3 more.

browser_typeType textC
Destructive

Type text into editable element

Inputs: text (required), target (required), _meta, slowly, +2 more.

browser_wait_forWait forC
Read-only

Wait for text to appear or disappear or a specified time to pass

Inputs: text, time, _meta, textGone.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelogโ€ฆ

    TDQS

    C2.9/5.0

    Scored across 24 tools

    Disambiguation4/5

    Most tools have clearly distinct purposes, but 'browser_snapshot' and 'browser_take_screenshot' could confuse agents as both capture page state, and 'browser_type' vs 'browser_fill_form' overlap slightly. Descriptions help differentiate them.

    Naming Consistency5/5

    All tools follow a consistent 'browser_<action_or_noun>' pattern using lowercase with underscores. The naming is predictable and uniform, making it easy for agents to infer tool function from the name.

    Tool Count5/5

    24 tools is well-scoped for a browser automation server, covering navigation, interaction, forms, dialogs, console, network, tabs, and more. Each tool addresses a distinct operation without being excessive.

    Completeness4/5

    The tool set covers most common browser automation operations (CRUD-like for page interactions), but lacks a forward navigation tool and explicit waiting for elements beyond text. These minor gaps are workable.

    Maintenance

    ActivityMaintained
    ResponsivenessLimited data

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      D
      maintenance
      An MCP server that enables AI-powered browser automation, web scraping, and testing using Playwright across Chromium, Firefox, and WebKit. It allows users to perform actions like navigation, clicking, typing, and taking screenshots through natural language interfaces.
      24 npm
      MIT
    • A
      license
      Not graded
      quality
      Not graded
      maintenance
      An MCP server that enables AI agents to automate browser interactions using Playwright and Cloudflare Workers, supporting tasks like navigation, clicking, typing, and screenshots.
      8,604 npm
      Apache 2.0
    • A
      license
      Not graded
      quality
      D
      maintenance
      A stealth-enhanced browser automation MCP server for AI agents to interact with websites while bypassing anti-bot detection mechanisms like Cloudflare and reCAPTCHA.
      9
      MIT