Selenium Boot MCP
ActiveA Model Context Protocol server that gives Claude and other AI assistants a real browser to drive — 84 tools for navigation, elements, assertions, and network control, with test-code generation for Java, Python, C#, and Gherkin.
The problem
Asking an AI assistant to “test the login flow” is easy to say and, without a real browser underneath it, impossible to actually do — the assistant can write plausible-looking Selenium code, but it can’t see the page, click the button, or tell you whether the assertion it wrote would even pass.
What it does
Selenium Boot MCP puts a real, driver-controlled Chrome session behind Claude, GitHub Copilot, or any MCP-speaking assistant, exposed as 84 tools across four groups: browser control (navigation, tabs, cookies, storage, network mocking, a built-in visual-regression diff, and a WCAG accessibility audit), element interaction (click, type, drag-and-drop, shadow DOM, iframes, self-healing locators), assertions, and code generation. Ask it to log in, click around, and assert a heading — then ask it to turn that session into a Java TestNG class, a JUnit 5 test, a Gherkin feature file with step definitions, a pytest class, or a C# NUnit test, plus the GitHub Actions or Jenkins pipeline to run it. Chrome starts automatically on first use; there’s no separate ChromeDriver to install or version-match.
Where it runs
Ships as a pip install-able PyPI package, wired up with one click via a VS
Code Marketplace extension (registers with GitHub Copilot and Claude Code) or
a JetBrains plugin (registers with the JetBrains AI Assistant) — both install
the Python package automatically if it’s missing. Claude Desktop works too,
with one block of JSON in its config.
Under the hood
Python driving Selenium WebDriver, implementing the Model Context Protocol so any compliant AI assistant can call its tools directly. MIT-licensed and open source, and — despite the name — not a dependency of the Selenium Boot framework: it drives any Selenium session, framework or none.