Open-source AI browser assistant — any model, anywhere.
Claude for Chrome is great but geo-restricted and locked to expensive Claude models.
SuperDuck fixes both — use any model, from anywhere:
| Claude for Chrome | SuperDuck | |
|---|---|---|
| Region | US/UK only | Global |
| Models | Claude only | Claude, DeepSeek, Qwen, Kimi, Gemini, GPT, or any OpenAI-compatible API |
| Cost | Fixed pricing | Your API key, your budget |
| Browser control | Limited | Full CDP: click, scroll, type, screenshot, JS exec |
| Extensibility | Closed | MCP runtime, workflow recording, CLI |
Multi-model routing — Map Opus / Sonnet / Haiku slots to any provider's model. Switch per task.
Browser automation — Screenshot, click, scroll, type, navigate, and execute JavaScript on any tab via CDP.
Workflow recording — Record browser actions with voice narration. AI replays them for repeatable automation.
MCP runtime — Built-in Model Context Protocol runtime. Connect to any MCP-compatible server.
CLI + MCP server — superduck CLI lets AI coding agents (Claude Code, Codex, Cursor) drive your live Chrome session. Same cookies, same login — no headless browser.
Refined UI — Markdown with syntax highlighting, LaTeX math, image attachments, dark/light themes.
Point Settings at any OpenAI-compatible endpoint, then assign each slot:
| Slot | Default | Custom example |
|---|---|---|
| Opus | claude-opus-4-6 |
deepseek-r1 / gpt-4o |
| Sonnet | claude-sonnet-4-6 |
qwen-max / kimi-k2.5 |
| Haiku | claude-haiku-4-5 |
deepseek-chat / gemini-2.5-flash |
npm install -g superduck-cli
superduck setup # install native host & permissions
superduck doctor # verify everything is wired- Download the latest release or build from source
- Go to
chrome://extensions/→ enable Developer mode - Click Load unpacked → select the
chrome-crx/dist/folder - Pin SuperDuck to your toolbar
The superduck CLI bridges AI coding agents to your live Chrome session via Unix domain socket + Chrome Native Messaging.
Agent CLI ──▶ superduck ──▶ native host ──▶ SuperDuck CRX ──▶ tab
claude code CLI binary Go process chrome extension live page
└── UDS ─┘ └── chrome.runtime ─┘
Key commands:
superduck context # read active tab url/title/selection
superduck screenshot # capture viewport
superduck --tab $TAB navigate https://example.com
superduck --tab $TAB left_click 120 240
superduck --tab $TAB type "hello"
superduck --tab $TAB exec "document.title"
superduck --tab $TAB read_page --filter interactive
superduck --tab $TAB console --pattern errorRun superduck --help for the full command reference.
git clone https://github.com/superduck-ai/superduck.git
cd superduck/chrome-crx
bun install
bun run build # Output → chrome-crx/dist/Development:
bun run dev # Watch mode
bun run typecheck # Type check
bun run lint # Lint
bun run test # Run testsReact 19 · TypeScript 5 · Vite 7 · Tailwind CSS 4 · Zustand · Framer Motion · Tiptap · @anthropic-ai/sdk · MCP SDK · Go
Contributions welcome! Fork → branch → commit → PR.
Made with 🦆 by the SuperDuck community