Camoufox MCP
<div align="center">
# π¦ Camoufox MCP
**Anti-detection browser automation for AI agents**
[](https://npmjs.org/package/camoufox-mcp)
[](https://npmjs.org/package/camoufox-mcp)
[](https://opensource.org/licenses/MIT)
[](https://nodejs.org/)
MCP server for browser automation using [Camoufox](https://camoufox.com/) - an anti-detection Firefox browser.
</div>
## Installation
```bash
npx camoufox-mcp
```
## Claude Desktop Configuration
Add to `claude_desktop_config.json`:
**Basic:**
```json
{
"mcpServers": {
"camoufox": {
"command": "npx",
"args": ["camoufox-mcp"]
}
}
}
```
**With Proxy (via env vars):**
```json
{
"mcpServers": {
"camoufox": {
"command": "npx",
"args": ["camoufox-mcp"],
"env": {
"PROXY_SERVER": "http://1.2.3.4:8080",
"PROXY_USER": "username",
"PROXY_PASS": "password"
}
}
}
}
```
Then in `~/.config/camoufox-mcp/settings.json`:
```json
{
"proxy": {
"enabled": true,
"server": "${PROXY_SERVER}",
"username": "${PROXY_USER}",
"password": "${PROXY_PASS}",
"geoip": true
}
}
```
## Why Camoufox MCP?
| Feature | Camoufox MCP | Chrome DevTools MCP |
|---------|--------------|---------------------|
| **Anti-Detection** | | |
| Bot detection bypass | β
Built-in | β Detected |
| Fingerprint randomization | β
| β |
| WebRTC leak protection | β
| β |
| Canvas fingerprint spoofing | β
| β |
| **Automation** | | |
| Humanized cursor movement | β
Realistic paths | β Instant jumps |
| Click with human-like delays | β
| β |
| Proxy with GeoIP auto-config | β
Timezone/locale | β Manual |
| **JavaScript Evaluation** | | |
| MainWorld eval (`__NUXT__`, `__NEXT_DATA__`) | β
| β
|
| Isolated eval (async/await) | β
| β
|
| **Session Management** | | |
| Isolated browser sessions | β
Multiple contexts | β Single context |
| Pre-authenticated sessions | β
Cookie/localStorage injection | β |
| Session state export | β
Save to JSON | β |
| **Features** | | |
| Cookie popup auto-dismiss | β
50+ selectors | β |
| User action recording | β
| β |
| Network interception | β
| β
|
| Console capture | β
| β
|
| Performance tracing | β | β
|
| **Browser** | | |
| Engine | Firefox (Gecko) | Chrome (Chromium) |
| Auto-install | β
~150MB | β
|
## Tools
### Session Management
Sessions provide isolated browser contexts β each session has its own cookies, localStorage, and cache. Use sessions to run multiple authenticated profiles simultaneously.
| Tool | Description |
|------|-------------|
| `create_session` | Create a new isolated session, optionally pre-loading cookies/localStorage from a JSON file |
| `list_sessions` | List all sessions with their page counts and metadata |
| `save_session` | Export a session's cookies and localStorage to a JSON file |
| `close_session` | Close a session and all its pages |
| `list_context_files` | List available context files registered in settings |
### Page Management
| Tool | Description |
|------|-------------|
| `new_page` | Create a new page, optionally in a specific session |
| `list_pages` | List all open pages with their session IDs |
| `select_page` | Set a page as active |
| `close_page` | Close a page |
### Navigation & Interaction
| Tool | Description |
|------|-------------|
| `navigate_page` | Navigate to URL, back, forward, or reload |
| `click` | Click an element by UID |
| `hover` | Hover over an element |
| `fill` | Fill a form field |
| `fill_form` | Fill multiple form fields at once |
| `press_key` | Press a key or key combination |
| `drag` | Drag an element to another |
### Inspection
| Tool | Description |
|------|-------------|
| `take_snapshot` | Text snapshot with interactive element UIDs |
| `take_screenshot` | Screenshot of page or element |
| `get_page_text` | Extract text content |
| `evaluate_mainworld` | Run JS in page context (sync only) |
| `evaluate_isolated` | Run JS in isolated context (async OK) |
| `get_framework_state` | Extract Next.js/Nuxt.js state |
### Context Injection
| Tool | Description |
|------|-------------|
| `inject_context` | Inject cookies + localStorage from a JSON file into a session |
### Monitoring
| Tool | Description |
|------|-------------|
| `start_network_capture` / `stop_network_capture` | Capture network requests |
| `get_network_logs` / `clear_network_logs` | View/clear captured requests |
| `intercept_network` / `remove_intercept` | Block or modify requests |
| `start_console_capture` / `stop_console_capture` | Capture console messages |
| `get_console_logs` / `clear_console_logs` | View/clear console messages |
| `start_recording` / `stop_recording` / `get_recording_status` | Record user interactions |
### Utility
| Tool | Description |
|------|-------------|
| `dismiss_popups` | Auto-dismiss cookie consent popups |
| `wait_for` | Wait for text or selector |
| `wait_for_navigation` | Wait for navigation to complete |
| `reload_settings` | Reload settings from config |
| `get_config_paths` | Show config file paths |
| `init_user_config` | Initialize user config directory |
## Sessions
Sessions enable running multiple authenticated profiles in parallel. Each session is an isolated browser context with its own cookies and localStorage.
```
Browser (single Camoufox instance)
βββ Session "github" (cookies from github.json)
β βββ Page: github.com
β βββ Page: github.com/repo
βββ Session "linkedin" (cookies from linkedin.json)
β βββ Page: linkedin.com/feed
βββ Session "default" (no pre-loaded state)
βββ Page: example.com
```
### Usage Examples
**Create a session with pre-loaded authentication:**
1. `create_session` with `contextPath: "github-cookies.json"` β returns a `sessionId`
2. `new_page` with `url: "https://github.com"` and the `sessionId` β opens GitHub already logged in
**Run two accounts simultaneously:**
1. `create_session` with `contextPath: "account1.json"` β session A
2. `create_session` with `contextPath: "account2.json"` β session B
3. `new_page` in session A β logged in as account 1
4. `new_page` in session B β logged in as account 2
**Save session state for later:**
1. Log in manually in a session
2. `save_session` with `outputPath: "my-session.json"` β exports cookies + localStorage
3. Next time, `create_session` with `contextPath: "my-session.json"` β restored
### Context File Format
Context files use [Playwright's storageState format](https://playwright.dev/docs/api/class-browsercontext#browser-context-storage-state):
```json
{
"cookies": [
{
"name": "session_id",
"value": "abc123",
"domain": ".example.com",
"path": "/",
"expires": 1735689600,
"httpOnly": true,
"secure": true,
"sameSite": "Lax"
}
],
"origins": [
{
"origin": "https://example.com",
"localStorage": [
{ "name": "token", "value": "xyz789" }
]
}
]
}
```
### Registering Context Files in Settings
Add context file paths to `settings.json` to make them discoverable via `list_context_files`. Sessions are created **lazily on demand** β no memory is used until the agent calls `create_session`:
```json
{
"contextPaths": [
"/path/to/github-cookies.json",
"/path/to/linkedin-cookies.json"
]
}
```
The agent can then call `list_context_files` to see what's available, and `create_session` with the desired `contextPath` to load it when needed.
## Settings
Create `~/.config/camoufox-mcp/settings.json`:
```json
{
"browser": {
"headless": false,
"viewport": [1280, 1080],
"timeout": 30000,
"humanize": 0.5
},
"proxy": {
"enabled": true,
"server": "http://proxy.example.com:8080",
"username": "user",
"password": "pass",
"geoip": true
},
"contextPaths": []
}
```
### Proxy Examples
**HTTP/HTTPS Proxy:**
```json
{ "proxy": { "enabled": true, "server": "http://1.2.3.4:8080" } }
```
**SOCKS5 Proxy:**
```json
{ "proxy": { "enabled": true, "server": "socks5://1.2.3.4:1080" } }
```
**With Authentication (using env vars):**
```json
{
"proxy": {
"enabled": true,
"server": "${PROXY_SERVER}",
"username": "${PROXY_USER}",
"password": "${PROXY_PASS}",
"geoip": true
}
}
```
> `geoip: true` auto-adjusts browser timezone/locale based on proxy IP location.
## See More
- [Camoufox](https://github.com/daijro/camoufox) - The anti-detection Firefox browser
- [camoufox-js](https://github.com/AnyFetch/camoufox-js) - JavaScript/TypeScript bindings
## License
MIT
TDQS
Scored across 36 tools
Most tools have distinct purposes, but there is some overlap that could cause confusion. For example, 'evaluate_isolated' and 'evaluate_mainworld' both execute JavaScript but in different contexts, which is clarified in descriptions, yet their similar names might lead to misselection. Tools like 'fill' and 'fill_form' also have overlapping functionality, though 'fill_form' handles multiple fields. Overall, descriptions help differentiate, but some boundaries are unclear.
Tool names largely follow a consistent verb_noun pattern (e.g., 'clear_console_logs', 'get_page_text', 'start_recording'), making them predictable and readable. There are minor deviations, such as 'evaluate_isolated' and 'evaluate_mainworld' using adjectives instead of nouns, and 'init_user_config' using 'init' instead of a more standard verb like 'initialize'. However, the overall naming scheme is coherent and easy to follow.
With 36 tools, the count is excessive for a browser automation server, leading to potential cognitive overload and reduced usability. A typical well-scoped server in this domain should have 10-20 tools; this set includes many granular operations (e.g., separate tools for starting/stopping console and network capture) that could be consolidated. The high number makes the tool surface feel heavy and less intuitive for agents.
The tool set provides comprehensive coverage for browser automation, including navigation, interaction (click, hover, drag), form handling, JavaScript execution, logging, network control, and state management. It supports full CRUD-like operations for logs and captures, and there are no obvious gapsβtools like 'wait_for' and 'wait_for_navigation' ensure robust workflow support. The domain is well-covered without dead ends.