selenium-mcp

Selenium MCP server for browser automation through Model Context Protocol clients.

It is built with TypeScript, the official MCP SDK, Selenium WebDriver, strict zod input validation, explicit waits, and structured responses that are useful to both humans and AI agents.

Features

• Launch Chrome, Firefox, or Edge

• Navigate pages and read URL/title/source

• Find, click, type, hover, right-click, double-click, and retry-click elements

• Read text and attributes

• Press keyboard keys

• Capture screenshots

• Upload files through file inputs

• Execute JavaScript

• Assert visible text, attributes, and visibility

• Manage tabs/windows, frames, alerts, and cookies

• Expose browser status and accessibility snapshot MCP resources

What's New In v0.1.0

• First public npm release: @gaforov/selenium-mcp

• Expanded toolset for end-to-end browser automation, assertions, and diagnostics

• Added MCP resources for browser status and accessibility snapshot context

• Added broad client setup documentation for IntelliJ, VS Code, Cursor, Claude, Goose, and Windsurf

• Added CI workflow and baseline tests for shared contracts

Requirements

• Node.js 20+

• Chrome, Firefox, or Edge installed

Selenium Manager handles browser drivers for supported local browsers.

Quick Start From Source

git clone https://github.com/gaforov/selenium-mcp.git
cd selenium-mcp
npm install
npm run build
npm start

MCP Client Config

Local source checkout:

{
  "mcpServers": {
    "selenium-mcp": {
      "command": "node",
      "args": ["/absolute/path/to/selenium-mcp/dist/server.js"]
    }
  }
}

After npm publish:

{
  "mcpServers": {
    "selenium-mcp": {
      "command": "npx",
      "args": ["-y", "@gaforov/selenium-mcp@latest"]
    }
  }
}

For IntelliJ IDEA, VS Code, Cursor, Claude Desktop, Claude Code, Goose, and Windsurf setup notes, see docs/CLIENT_INTEGRATION.md.

First Prompt To Try

Ask your MCP-capable assistant:

Use selenium-mcp to open Chrome, navigate to https://example.com, read the page title, take a screenshot, and close the browser.

Expected tool flow:

1. start_browser

2. navigate

3. get_title

4. take_screenshot

5. stop_browser

Implemented Tools

• start_browser

• open_url

• navigate

• find_element

• wait_for_element

• wait_until_visible

• click

• retry_click

• interact

• type

• get_text

• get_attribute

• assert_text

• assert_visible

• assert_attribute

• press_key

• take_screenshot

• get_current_url

• get_title

• get_page_source

• execute_script

• upload_file

• window

• frame

• alert

• add_cookie

• get_cookies

• delete_cookie

• stop_browser

MCP Resources

• browser-status://current

• accessibility://current

Documentation

• Tool reference

• Client integration

• Architecture

• Development guide

• Changelog

Development

npm run typecheck
npm run build
npm test

License

MIT. See LICENSE.