mcp-browser-automation

This MCP server enables browser automation using Playwright controlled by Claude Desktop prompts. The server exposes a Playwright-based automation endpoint that Claude Desktop can trigger through the MCP protocol. With this server running, you can prompt the Claude Desktop client to open a browser, navigate to URLs, fill forms, click elements, and perform scripted interactions in a browser session. Typical workflows include launching a browser, visiting a target page, extracting information, and returning results back to the Claude Desktop client. The server is designed to be invoked by the MCP client as part of a conversation, where each command from Claude translates into a browser action managed by Playwright.

Prerequisites:

• Node.js installed on your machine
• Claude Desktop installed (for triggering MCP commands)

Installation steps:

1. Clone the repository: git clone https://github.com/hrmeetsingh/mcp-browser-automation.git
2. Navigate to the project directory: cd mcp-browser-automation
3. Install dependencies: npm install
4. Build the project (if a build script is provided): npm run build
   • Ensure the dist/index.js file exists after the build.
5. Run the server (example): node dist/index.js
6. Verify the server is reachable by configuring Claude Desktop to point to the local MCP server path as shown in the integration steps of the README.

Tips and notes:

• Ensure Playwright browsers are installed (Playwright usually sets this up during npm install; you can run npx playwright install to be safe).
• In Claude Desktop, configure your MCP server with the path to the built dist/index.js as shown in the integration example.
• If you change the server entry file path, update the command args in the MCP configuration accordingly.
• Common issues include mismatched Node versions or missing dependencies; rerun npm install to fix missing modules.
• For macOS, the Claude Desktop config path commonly used is ~/Library/Application Support/Claude/claude_desktop_config.json; ensure the JSON structure matches the example.
• When testing, include explicit actions in your prompts (e.g., navigate to URL, wait for page load, locate element by selector, click, type text) to make automation deterministic.