web-eval-agent MCP Server

What is web-eval-agent?

web-eval-agent is an MCP server from operativelabs.xyz that launches a browser-use powered agent to autonomously execute and debug web apps directly inside your code editor. It is designed for developers who want to let a coding agent test its own code end-to-end while they focus on other tasks.

How to use web-eval-agent?

Obtain a free API key from operativelabs.xyz, then run the provided installer script (macOS/Linux) which installs Playwright and uv and inserts the required JSON configuration into your code editor (Cursor/Cline/Windsurf). After restarting the editor, send a chat prompt such as “Test my app on http://localhost:3000. Use web-eval-agent.” to invoke the tool. Manual installation steps are also documented for macOS/Linux and Windows.

Key features of web-eval-agent

• Navigates your webapp using BrowserUse (2× faster with operative backend)
• Captures network traffic with intelligently filtered requests
• Collects all console logs and errors
• Autonomous debugging—the Cursor agent calls web-eval-agent to test its own code
• Provides a structured report including agent steps, logs, network requests, and a timeline

Use cases of web-eval-agent

• Automatically testing a locally running web application after writing new code
• End-to-end validation of user flows (e.g., API key creation and deletion)
• Debugging interactive UI features without manually clicking through the browser
• Integrating autonomous QA directly into an LLM-powered coding agent’s workflow

FAQ from web-eval-agent

What prerequisites are needed to install web-eval-agent?

On macOS/Linux you may need Homebrew, npm, and jq, though typically these are not required. The installer will set up Playwright and the uv package manager.

How do I update web-eval-agent to the latest version?

Run uv cache clean and then refresh the MCP server in your code editor. To force a reinstall, use the installation script again.

Where does the agent run and where are logs stored?

The agent runs in a browser launched by Playwright on your local machine. A report with screenshots, console logs, network requests, and a timeline is returned into the editor context. Live logs are also available on the Operative Control Center dashboard.

What platforms and editors are supported?

The server works on macOS, Linux, and Windows. It integrates with Cursor, Cline, and Windsurf via MCP JSON configuration. Windows installation is still being refined—users may encounter issues.

Does web-eval-agent require an API key?

Yes, a free API key from operativelabs.xyz is required and must be set in the OPERATIVE_API_KEY environment variable in the editor’s MCP configuration.