OmniBrowser

The fastest multi-profile, multi-browser automation for AI agents. Works with Claude, GPT, Kimi K2, DeepSeek, Ollama. Supports Chrome, Edge, Arc, Brave, Dia.

The Problem

Claude in Chrome is great, but:

It only works with ONE profile - If you're a consultant with multiple client accounts, you're screwed
It only works with Chrome - Arc, Brave, Edge, Dia users are locked out
It's slow - Screenshot-analyze-act cycle adds latency on every action
It requires a subscription - API billing users can't use it

Real User Pain (from GitHub issues)

"The extension is bound to a single Chrome profile. Each automation attempt creates new windows in the WRONG profile instead of the active one." — #19740

"When running multiple Chrome instances, the Native Messaging API connects to whichever Chrome process responds first, leading to unpredictable behavior." — #15125

"Users resort to manual file export/import, defeating the purpose of browser automation." — #19740

The Solution

# Install
pip install omnibrowser

# Use with any profile
omnibrowser --profile "Client A" --browser chrome

# Or use the MCP server with Claude Code
claude --mcp omnibrowser

Commands

/profiles                    # List all available browser profiles
/profile "Work"              # Switch to Work profile
/profile "Client A"          # Switch to Client A profile
/incognito                   # Start a clean incognito session
/browser arc                 # Switch to Arc browser
/browser brave               # Switch to Brave browser

Features

1. Multi-Profile Support

Switch between Chrome profiles seamlessly. No more wrong-profile nightmares.

┌─────────────────┐     ┌─────────────────┐     ┌─────────────────┐
│ Chrome Profile  │     │ Chrome Profile  │     │ Chrome Profile  │
│ "Work"          │     │ "Client A"      │     │ "Client B"      │
└────────┬────────┘     └────────┬────────┘     └────────┬────────┘
         │                       │                       │
         └───────────────────────┼───────────────────────┘
                                 │
                                 ▼
                    ┌─────────────────────────┐
                    │     OmniBrowser MCP     │
                    │  /profile "Client A"   │
                    └─────────────────────────┘

How it works:

Enumerates all Chrome profiles from ~/Library/Application Support/Google/Chrome/
Uses DevTools Protocol (CDP) for targeted instance connections
Registers Native Messaging hosts per-profile for isolation

2. Multi-Browser Support

One MCP server, all Chromium browsers.

Browser	Status	Notes
Chrome	✅ Supported	Baseline
Edge	✅ Supported	Same Chromium APIs
Arc	✅ Supported	Power user favorite
Brave	✅ Supported	Privacy-focused crowd
Dia	✅ Supported	First & only Dia automation tool

Native Messaging Paths (macOS):

Chrome: ~/Library/Application Support/Google/Chrome/NativeMessagingHosts/
Edge:   ~/Library/Application Support/Microsoft Edge/NativeMessagingHosts/
Arc:    ~/Library/Application Support/Arc/User Data/NativeMessagingHosts/
Brave:  ~/Library/Application Support/BraveSoftware/Brave-Browser/NativeMessagingHosts/
Dia:    ~/Library/Application Support/Dia/NativeMessagingHosts/

3. Blazing Fast Mode

Skip screenshots when you don't need them. DOM-first approach for text-based tasks.

Mode	Speed	Use Case
Fast (DOM-only)	~50ms/action	Form filling, clicking, text extraction
Visual (screenshots)	~500ms/action	Complex UIs, visual verification
Hybrid	Auto-detect	Best of both worlds

Additional speed features:

Batch operations - Click 5 elements in 1 API call
Parallel tabs - Run actions across multiple tabs simultaneously
Skills system - Record a workflow once, replay via API (50x faster)

4. Model Agnostic

Works with any LLM. Your choice.

Provider	Models	Auth
Anthropic	Claude 3.5/4, Opus	API key or Max subscription (OAuth)
OpenAI	GPT-4, GPT-4o, GPT-5	API key
Moonshot	Kimi K2, K2.5	API key (cheap!)
DeepSeek	DeepSeek-V3, R1	API key
Ollama	Llama, Mistral, etc.	Local

Installation

Quick Start

pip install omnibrowser

As MCP Server (for Claude Code)

# Add to your Claude Code config
claude mcp add omnibrowser

# Or manually in ~/.claude/settings.json
{
  "mcpServers": {
    "omnibrowser": {
      "command": "omnibrowser",
      "args": ["--mcp"]
    }
  }
}

Browser Extension

Install the companion extension for each browser you want to use:

Chrome Extension (or Chrome Web Store link)
Edge Extension
Arc Extension
Brave Extension
Dia Skill

Usage

Basic Automation

from omnibrowser import OmniBrowser

browser = OmniBrowser(
    browser="chrome",
    profile="Work",
    model="claude"  # or "gpt4", "kimi", "ollama"
)

# Navigate and interact
browser.goto("https://docs.google.com")
browser.click("New Document")
browser.type("Meeting notes for Client A...")

MCP Tools

When used as an MCP server, OmniBrowser exposes these tools:

Tool	Description
`profiles_list`	List all available browser profiles
`profile_switch`	Switch to a specific profile
`browser_switch`	Switch to a different browser
`navigate`	Go to URL
`click`	Click element (by selector, text, or coordinates)
`type`	Type text into focused element
`screenshot`	Take screenshot
`read_page`	Get DOM/accessibility tree
`execute_js`	Run JavaScript
`batch_actions`	Execute multiple actions in one call
`skill_record`	Start recording a reusable skill
`skill_replay`	Replay a recorded skill

Skills System (Teach Once, Replay Fast)

# Record a login workflow
browser.skill_record("github_login")
browser.goto("https://github.com/login")
browser.type("#login_field", "username")
browser.type("#password", "password")
browser.click("Sign in")
browser.skill_stop()

# Replay it 50x faster (uses captured API calls, not UI)
browser.skill_replay("github_login")

Architecture

┌─────────────────────────────────────────────────────────────────┐
│                        OmniBrowser MCP                          │
├─────────────────────────────────────────────────────────────────┤
│  Profile Manager                                                 │
│  ├── Profile enumeration (reads browser user data dirs)         │
│  ├── Profile switching (CDP target discovery)                   │
│  └── Incognito session management                               │
├─────────────────────────────────────────────────────────────────┤
│  Browser Adapters                                                │
│  ├── ChromeAdapter    → Native Messaging + CDP                  │
│  ├── EdgeAdapter      → Native Messaging + CDP                  │
│  ├── ArcAdapter       → Native Messaging + CDP                  │
│  ├── BraveAdapter     → Native Messaging + CDP                  │
│  └── DiaAdapter       → Native Messaging + Dia Skill API        │
├─────────────────────────────────────────────────────────────────┤
│  Speed Layer                                                     │
│  ├── DOM-first mode (skip screenshots)                          │
│  ├── Batch action executor                                      │
│  ├── Parallel tab manager                                       │
│  └── Skills engine (macro recording + API replay)               │
├─────────────────────────────────────────────────────────────────┤
│  Model Router                                                    │
│  ├── Claude (API + OAuth for Max subscription)                  │
│  ├── OpenAI                                                     │
│  ├── Kimi K2                                                    │
│  ├── DeepSeek                                                   │
│  └── Ollama (local)                                             │
└─────────────────────────────────────────────────────────────────┘

Comparison

Feature	OmniBrowser	Claude in Chrome	browser-use	Playwright
Multi-profile	✅	❌	❌	❌
Multi-browser	✅ Chrome/Edge/Arc/Brave/Dia	❌ Chrome only	✅ Chromium	✅ All
AI-native	✅	✅	✅	❌
Model-agnostic	✅ Any LLM	❌ Claude only	✅ Any LLM	N/A
Speed mode	✅ DOM-first	❌ Screenshot-based	❌ Screenshot-based	✅ Fast
Skills/macros	✅	❌	✅	❌
Uses your logins	✅	✅	❌	❌
MCP server	✅	✅	✅	❌

Configuration

Environment Variables

OMNIBROWSER_DEFAULT_BROWSER=chrome    # Default browser
OMNIBROWSER_DEFAULT_PROFILE=Default   # Default profile
OMNIBROWSER_MODEL=claude              # Default LLM
OMNIBROWSER_SPEED_MODE=hybrid         # fast|visual|hybrid

Config File

~/.omnibrowser/config.json:

{
  "defaultBrowser": "chrome",
  "defaultProfile": "Work",
  "model": {
    "provider": "anthropic",
    "model": "claude-sonnet-4-20250514",
    "apiKey": "${ANTHROPIC_API_KEY}"
  },
  "speedMode": "hybrid",
  "skills": {
    "directory": "~/.omnibrowser/skills"
  }
}

Roadmap

v0.1 (MVP)

Multi-profile support for Chrome
Basic MCP tools (navigate, click, type, screenshot)
Claude API integration

v0.2

Multi-browser support (Edge, Arc, Brave)
DOM-first speed mode
OpenAI + Kimi K2 support

v0.3

Skills system (record/replay)
Batch operations
Parallel tab execution

v0.4

Dia browser support (first-ever!)
Ollama local model support
Chrome extension for easy setup

v1.0

Production-ready stability
Full documentation
Chrome Web Store / extension marketplaces

Contributing

Contributions welcome! See CONTRIBUTING.md.

Development Setup

git clone https://github.com/yourusername/omnibrowser
cd omnibrowser
python -m venv venv
source venv/bin/activate
pip install -e ".[dev]"

Running Tests

pytest tests/

Why "OmniBrowser"?

Omni = all, every
Works with all profiles
Works with all Chromium browsers
Works with all major LLMs

License

MIT License. See LICENSE.

Acknowledgments

Inspired by Claude in Chrome by Anthropic
browser-use for proving AI browser automation at scale
The frustrated users on GitHub issues who made the problems crystal clear

Built with frustration, shipped with love.
Because switching Chrome profiles shouldn't require a PhD.

Name		Name	Last commit message	Last commit date
Latest commit History 3 Commits
docs		docs
extensions/chrome		extensions/chrome
scripts		scripts
src/omnibrowser		src/omnibrowser
.gitignore		.gitignore
CONTRIBUTING.md		CONTRIBUTING.md
LICENSE		LICENSE
README.md		README.md
pyproject.toml		pyproject.toml
test_screenshot_Work.png		test_screenshot_Work.png
test_screenshot_simple.png		test_screenshot_simple.png

Folders and files

Latest commit

History

Repository files navigation

OmniBrowser

The Problem

Real User Pain (from GitHub issues)

The Solution

Commands

Features

1. Multi-Profile Support

2. Multi-Browser Support

3. Blazing Fast Mode

4. Model Agnostic

Installation

Quick Start

As MCP Server (for Claude Code)

Browser Extension

Usage

Basic Automation

MCP Tools

Skills System (Teach Once, Replay Fast)

Architecture

Comparison

Configuration

Environment Variables

Config File

Roadmap

v0.1 (MVP)

v0.2

v0.3

v0.4

v1.0

Contributing

Development Setup

Running Tests

Why "OmniBrowser"?

License

Acknowledgments

About

Resources

License

Contributing

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages