Selenium MCP

What seleniumboot/selenium-mcp MCP server does

seleniumboot/selenium-mcp MCP server lets an MCP client control a Selenium WebDriver session through browser, element, assertion, storage, diagnostics, and code-generation tools. It is intended for tasks such as opening a web page, filling forms, clicking controls, checking visible content, and recording the actions performed during a session.

The server works with a real browser rather than only returning HTTP responses. Chrome is supported by the documented installation path, and the server uses Selenium Manager so a separate ChromeDriver installation is not required. The optional browser-start tool can select Firefox, enable headless mode, or set a custom window size.

How it works

Install the Python package and register the seleniumboot-mcp executable as an MCP server using a supported client configuration. Browser startup is lazy: the first browser-related tool call starts Chrome when no session is already open. Calling start_browser explicitly is useful when browser type, headless operation, or window dimensions must be selected before other actions.

The session can span navigation, multiple tabs or windows, page scrolling, JavaScript execution, cookies, local storage, session storage, frames, alerts, shadow DOM elements, and file uploads. inspect_page reports interactive controls and suggested CSS selectors, which can help an agent choose locators before acting. The server also records actions and supports viewing or clearing the session log.

Setup and configuration

The README documents Python 3.10 or newer and Chrome as requirements. Install the package with:

pip install seleniumboot-mcp

For Claude Desktop, add an MCP entry whose command is seleniumboot-mcp, then restart Claude Desktop. The documented configuration locations are %APPDATA%\\claude-desktop\\config.json on Windows and ~/.config/claude-desktop/config.json on macOS.

VS Code users can install the Selenium Boot MCP Marketplace extension, which registers the server with GitHub Copilot and can create project MCP configuration. JetBrains users can install the corresponding Marketplace plugin for JetBrains AI Assistant. Manual VS Code and Claude Code configurations are also documented in the repository.

Tools and capabilities

seleniumboot/selenium-mcp MCP server includes tools for:

• Browser navigation, history, tabs, screenshots, page source, titles, URLs, scrolling, device emulation, and JavaScript.
• Element discovery and interaction, including clicks, typing, hovering, drag-and-drop, selects, waits, uploads, alerts, frames, shadow DOM, and tables.
• Assertions for titles, URLs, text, visibility, attributes, page content, and element counts.
• Browser state and diagnostics, including console logs, network logs, cookies, storage, response mocking, screenshot comparison, and accessibility checks.
• Self-healed locator inspection and cache management.
• Test generation for Java TestNG, Java JUnit 5, Java Page Objects, Gherkin with Java steps, pytest, C# NUnit, and equivalent Playwright TypeScript.
• CI output for GitHub Actions, Jenkins, and GitLab CI using Maven, Gradle, or pytest options where supported by the generator.

Limitations and notes

Chrome-only behavior applies to console logs and network logs according to the tool documentation. Mobile emulation uses Chrome DevTools Protocol. The server operates against a live browser session, so browser state, authentication, application data, and any actions requested by the MCP client affect that session. The provided material does not document environment variables, remote hosting, or API-key authentication.