What will this quick-start pack do for you?
Let OpenCode control browsers and desktop applications:
- Install Playwright MCP (web control)
- Install open-computer-use (desktop automation)
- Write the opencode.json configuration
- Verify the connection
Prerequisites
- Node.js 18+ is installed
- The computer has internet access
Ask OpenCode to run the following steps for me
Step 1: Write the Playwright MCP configuration
Edit ~/.config/opencode/opencode.json, add:
{
"mcp": {
"playwright": {
"type": "local",
"command": ["npx", "-y", "@playwright/mcp"],
"enabled": true
}
}
}
Step 2: Install open-computer-use
npm install -g open-computer-use
Confirm the command is available:
open-computer-use --version
Step 3: Write the open-computer-use MCP configuration
⚠️ Windows note:
npx open-computer-use install-opencode-mcpwill promptrequires a POSIX shell. On Windows, edit the configuration file manually.
Edit ~/.config/opencode/opencode.json, add:
{
"mcp": {
"open-computer-use": {
"type": "local",
"command": ["open-computer-use", "mcp"],
"enabled": true
}
}
}
💡 Why not use npx? The MCP server of open-computer-use needs to be started with the
open-computer-use mcpsubcommand. Usingnpx -y open-computer-usewill not correctly expose the MCP tools.
Step 3: Verify Playwright
Reopen OpenCode, then ask it:
請開啟 https://example.com 並告訴我頁面標題是什麼。
Step 4: Verify open-computer-use
請截圖我的桌面(或特定視窗)。
Completion report format
## 瀏覽器控制安裝完成
- Playwright MCP:已設定 / 已啟用
- open-computer-use:已設定 / 已啟用
- 瀏覽器導航測試:成功 / 失敗
- 桌面截圖測試:成功 / 失敗
Lightweight alternative: CLI-Anything Browser CLI
If you only need simple web operations (navigation, screenshots, content extraction) and do not need full Playwright or desktop automation, you can use CLI-Anything's Browser CLI:
Installation
pip install cli-anything-hub
cli-hub install browser
Usage example
cli-anything-browser navigate https://example.com
cli-anything-browser screenshot --output page.png
cli-anything-browser extract --selector "main"
💡 Playwright MCP (the default option in this quick-start pack) is suitable for full browser automation and form filling; Browser CLI is suitable for fast, lightweight web data extraction. The two can coexist.
FAQ
| Problem | Solution |
|---|---|
| Playwright is very slow when downloading the browser on first run | Wait for the download to finish; it is only needed once |
| open-computer-use cannot find the window | Confirm the target application is running |
npx open-computer-use tools do not appear |
Switch to command: ["open-computer-use", "mcp"] |
Windows install-opencode-mcp reports an error |
Manually edit opencode.json; do not run the automatic installation |