Skip to main content

kogiqa-mcp-browser-automation

MCP server for browser automation using natural language through kogiQA, enabling AI agents to interact with web pages without selectors

跳到安装

来源信息

仓库
reason-machines/mcp-skills
最近来源活动
2026年7月8日 07:19
检测到的 SKILL.md 语言
英语
星标
7
分支
2

安装方式

默认使用会先检查来源的 Prompt;你也可以切换为直接命令,或下载本地副本。

检查来源文件

决定是否安装前,请先阅读 SKILL.md,以及 SkillsMP 当前展示的配套文件。

正在显示 SKILL.md

SKILL.md
来源说明 · 只读预览
name
kogiqa-mcp-browser-automation
description
MCP server for browser automation using natural language through kogiQA, enabling AI agents to interact with web pages without selectors
triggers
["automate a web browser","interact with a webpage using kogiqa","test a web application with browser automation","navigate and click elements on a page","fill out web forms automatically","scrape content from a website","debug web applications with browser control","automate web testing tasks"]
# kogiQA MCP Browser Automation > Skill by [ara.so](https://ara.so) — MCP Skills collection. ## Overview kogiQA MCP is a Model Context Protocol server that provides browser automation capabilities using the kogiQA browser control algorithm. Unlike traditional automation tools, it enables AI agents to interact with web pages through natural language commands without requiring CSS selectors or XPath expressions. This approach saves tokens and simplifies web automation workflows. **Key advantages:** - No selector required — interact with elements using natural language descriptions - Built on Playwright for robust browser automation - Works with any MCP-compatible client (Claude Desktop, Cursor, VS Code, etc.) - Lightweight and easy to install ## Installation ### Quick Install The fastest way to install is using npx: ```bash npx kogiqa-mcp@latest ``` ### Claude Code ```bash claude mcp add kogiqa-browser npx kogiqa-mcp@latest ``` ### VS Code / VS Code Insiders Install via CLI: ```bash code --add-mcp '{"name":"kogiqa-browser","command":"npx","args":["kogiqa-mcp@latest"]}' ``` Or add manually to your MCP settings (`settings.json`): ```json { "mcp.servers": { "kogiqa-browser": { "command": "npx", "args": ["kogiqa-mcp@latest"] } } } ``` ### Cursor Add to Cursor Settings → MCP → Add new MCP Server: ```json { "kogiqa-browser": { "command": "npx", "args": ["kogiqa-mcp@latest"] } } ``` ### Standard MCP Configuration For Claude Desktop, Windsurf, Goose, or other MCP clients, add to your MCP configuration file: ```json { "mcpServers": { "kogiqa-browser": { "command": "npx", "args": ["kogiqa-mcp@latest"] } } } ``` Configuration file locations: - **Claude Desktop (macOS):** `~/Library/Application Support/Claude/claude_desktop_config.json` - **Claude Desktop (Windows):** `%APPDATA%\Claude\claude_desktop_config.json` - **Cursor:** `.cursor/config.json` in your project or user settings - **VS Code:** User or workspace settings ## Available Tools Once installed, the MCP server provides several tools for browser automation: ### `kogiqa_navigate` Navigate to a URL. **Parameters:** - `url` (string, required): The URL to navigate to **Example usage:** ```javascript // AI agent will call this tool as: kogiqa_navigate({ url: "https://example.com" }) ``` ### `kogiqa_click` Click on an element using natural language description. **Parameters:** - `description` (string, required): Natural language description of the element to click **Example usage:** ```javascript // No selectors needed - describe the element naturally kogiqa_click({ description: "the blue login button" }) kogiqa_click({ description: "submit button at the bottom" }) kogiqa_click({ description: "first link in the navigation menu" }) ``` ### `kogiqa_type` Type text into an input field. **Parameters:** - `description` (string, required): Natural language description of the input field - `text` (string, required): Text to type **Example usage:** ```javascript kogiqa_type({ description: "email input field", text: "user@example.com" }) kogiqa_type({ description: "search box at the top", text: "kogiQA MCP" }) ``` ### `kogiqa_get_content` Extract content from the current page. **Returns:** HTML or text content of the page **Example usage:** ```javascript kogiqa_get_content() ``` ### `kogiqa_screenshot` Take a screenshot of the current page. **Returns:** Base64-encoded PNG image **Example usage:** ```javascript kogiqa_screenshot() ``` ### `kogiqa_close` Close the browser instance. **Example usage:** ```javascript kogiqa_close() ``` ## Common Patterns ### Login Flow Automation ```javascript // Navigate to login page await kogiqa_navigate({ url: "https://app.example.com/login" }); // Fill in credentials await kogiqa_type({ description: "username field", text: process.env.USERNAME }); await kogiqa_type({ description: "password field", text: process.env.PASSWORD }); // Submit form await kogiqa_click({ description: "login button" }); // Verify success const content = await kogiqa_get_content(); ``` ### Form Submission ```javascript // Navigate to form await kogiqa_navigate({ url: "https://example.com/contact" }); // Fill multiple fields await kogiqa_type({ description: "name input", text: "John Doe" }); await kogiqa_type({ description: "email input", text: "john@example.com" }); await kogiqa_type({ description: "message textarea", text: "Hello!" }); // Submit await kogiqa_click({ description: "send button" }); ``` ### Web Scraping ```javascript // Navigate to target page await kogiqa_navigate({ url: "https://example.com/products" }); // Extract content const content = await kogiqa_get_content(); // Take screenshot for verification const screenshot = await kogiqa_screenshot(); // Parse content (this would be done by the AI agent) // Extract product names, prices, etc. ``` ### Multi-Step Navigation ```javascript // Start at homepage await kogiqa_navigate({ url: "https://example.com" }); // Navigate through site await kogiqa_click({ description: "products menu item" }); await kogiqa_click({ description: "first product card" }); await kogiqa_click({ description: "add to cart button" }); await kogiqa_click({ description: "checkout button" }); // Capture final state const screenshot = await kogiqa_screenshot(); ``` ### Testing with Verification ```javascript // Navigate to page under test await kogiqa_navigate({ url: "https://app.example.com/dashboard" }); // Take before screenshot const before = await kogiqa_screenshot(); // Perform action await kogiqa_click({ description: "settings button" }); // Verify change const content = await kogiqa_get_content(); const after = await kogiqa_screenshot(); // Close when done await kogiqa_close(); ``` ## Configuration Options ### Headless Mode By default, kogiQA MCP runs in headless mode. To run with a visible browser (useful for debugging), set environment variables: ```json { "mcpServers": { "kogiqa-browser": { "command": "npx", "args": ["kogiqa-mcp@latest"], "env": { "HEADLESS": "false" } } } } ``` ### Browser Type Specify browser (chromium, firefox, webkit): ```json { "mcpServers": { "kogiqa-browser": { "command": "npx", "args": ["kogiqa-mcp@latest"], "env": { "BROWSER_TYPE": "firefox" } } } } ``` ## Troubleshooting ### Server Not Starting **Issue:** MCP server fails to start or connect **Solutions:** - Ensure Node.js 20+ is installed: `node --version` - Clear npm cache: `npm cache clean --force` - Try installing globally: `npm install -g kogiqa-mcp@latest` - Check MCP configuration syntax in your client's config file ### Browser Launch Failures **Issue:** Browser fails to launch or crashes **Solutions:** - Install Playwright browsers: `npx playwright install` - Check for system dependencies: `npx playwright install-deps` - Try different browser: Set `BROWSER_TYPE=firefox` or `webkit` - Disable headless mode for debugging: `HEADLESS=false` ### Element Not Found **Issue:** kogiqa cannot find described element **Solutions:** - Be more specific in element descriptions: "blue submit button at bottom right" - Include context: "search input in the header" - Wait for page to load: Get content first to verify page state - Take screenshot to see current page state ### Timeouts **Issue:** Operations timeout waiting for elements **Solutions:** - Ensure page has fully loaded before interacting - Use more specific element descriptions - Check if element is hidden or covered by other elements - Verify the element actually exists using `kogiqa_get_content` ### Permission Issues **Issue:** Browser automation blocked by website **Solutions:** - Some sites block automation; this is expected behavior - Check site's terms of service for automation policies - Consider adding delays between actions - Use stealth mode if available in future versions ## Best Practices 1. **Descriptive element references:** Be specific but natural ("red delete button in the sidebar" not just "button") 2. **Verify page state:** Use `kogiqa_get_content` to confirm page loaded before interacting 3. **Handle secrets properly:** Use environment variables, never hardcode credentials 4. **Close browsers:** Call `kogiqa_close` when done to free resources 5. **Progressive navigation:** Navigate step-by-step rather than assuming page state 6. **Take screenshots:** Use for debugging and verification of automation flows 7. **Natural language:** Write descriptions as you would explain to a human ## Resources - **npm package:** https://www.npmjs.com/package/kogiqa-mcp - **GitHub repository:** atagon-GmbH/kogiqa-mcp - **License:** MIT
在 GitHub 查看