Chisle
코딩 에이전트 플러그인, 20개 실시간 작업에서 기본 모델 52% 활용, 간결한 페르소나와 출력 압축 제공
저장소 정보는 아직 수집하지 않았습니다.
아직 감지된 업데이트가 없습니다. GitHub 릴리스가 나오면 여기에 보입니다.
A private, agentic web browser for macOS on Chromium. Gives AI agents like Claude Code a better browser than Playwright.
README 전체를 불러오는 중입니다.
README 전문 보기 ↗Wink is a native Mac browser (AppKit and SwiftUI) on the Chromium engine. Agents are first-class: a built-in MCP server lets Claude Code, or any MCP client, browse with Wink, in fewer steps than Playwright. You see every step, and Wink asks before anything consequential.
1. Install. Download Wink (signed and notarized), drag it to Applications and open it once. If Claude Code is on your Mac, Wink adds itself to Claude Code's MCP servers. Nothing opens until an agent uses a tool; then Wink opens by itself.
2. Ask. In any Claude Code session:
Open localhost:3000 in Wink, sign up with a test account, and tell me what errors show up in the console.
The first time, Wink asks whether to let Claude Code browse.
3. Go headless (optional). For parallel sessions or CI, give each agent a private, invisible, throwaway Wink:
claude mcp add --scope user wink-headless -- /Applications/Wink.app/Contents/MacOS/wink-mcp --headless
claude mcp add --scope user wink -- /Applications/Wink.app/Contents/MacOS/wink-mcp
Any client that speaks MCP over stdio works the same way:
{ "mcpServers": { "wink": { "command": "/Applications/Wink.app/Contents/MacOS/wink-mcp" } } }
To stop Wink adding itself, turn it off in Settings › Agents or run claude mcp remove wink; it won't add itself again.
Same agent, same prompts, both browsers headless and private. Claude Sonnet 5.5 ran eleven browser tasks three times in each. Wink 0.8.1 was against Playwright MCP 0.0.85, with Playwright's network, storage, PDF and DevTools options turned on.
| Tool calls | Time | Input tokens | Tasks done | |
|---|---|---|---|---|
| Wink | 21% fewer | 28% less | 46% fewer | 33 / 33 |
| Playwright MCP | 33 / 33 |
Wink was faster on each of the eleven tasks and never needed more steps. Two more comparisons:
When Playwright is still the tool: Wink runs on Macs, on Chromium. For test suites on Linux or Windows servers, in Firefox's and Safari's own engines, or headless by the hundred on a server farm, use Playwright. Wink's export_test writes the Playwright tests those servers run, so you can use both. Full method and every number: wink.minnebo.ai/vs-playwright.
Each of the 37 tools is built to save a model turn. find returns just the matching controls in milliseconds. Each action replies with what changed: dialogs that opened, controls that changed state, text that updated. batch runs up to 40 steps across page loads, by CSS selector, visible text or field label.
| Tools | |
|---|---|
| Tabs | tabs_list get_active_tab tab_open tab_close tab_focus |
| Navigate | navigate go_back go_forward reload wait_for |
| Look | find snapshot read_page screenshot get_selection |
| Act | click type fill_form select_option press_key hover scroll drag upload_file handle_dialog batch |
| Develop | console_logs network_requests evaluate emulate mock_network storage |
| Keep | verify export_test trace record save_pdf |
Web developers get a workbench on localhost: requests in full with headers and bodies, mocks and offline mode, cookies and storage, phone sizes and dark mode, and JavaScript without prompts. export_test turns an agent's exploration into a Playwright test, and Wink's own CI runs those tests in Chromium, Firefox and WebKit. Details in docs/AGENTS.md.
여기까지가 README 앞부분입니다. 나머지는 원문에서 읽을 수 있습니다. README 전문 보기 ↗