M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
S

Selenium MCP Server

Community
MCP server for Selenium WebDriver – browser automation for AI agents.
How to install ↓ Compare GitHub source repository ↗
Category
Browser #17 of 32
Stars
★ 428 Popular
Transport
stdio (local process)
Runtime
Node.js
Credentials
No credential needed
License
MIT
Last commit
⚠
Tools
18
52FMRS · D

This MCP server provides a comprehensive set of browser automation tools, supporting major browsers, and can be invoked by AI agents via natural language. It's community-maintained, not official, but well-documented and easy to install. Best for developers and testers; be aware of potential security risks and website detection. No official client list is provided, but generic configuration is supported.

Strongest · Documentation 16/20 Weakest · Reliability 6/20

Reliability
6/20
Security and permissions
8/20
Maintenance
8/20
Documentation
16/20
Setup experience
14/20
Why each score
Reliability 6/20
Only the README and repository metadata are available. The README lists 16 tools with coherent parameters, but no source, package.json, CI, or tests were provided to verify the MCP handshake or actual tool behavior. Static calibration caps reliability at 12 and execution evidence is absent, so a conservative 6 is given.
Security and permissions 8/20
No red-line issues such as malware, credential theft, or real tokens in install examples were found. However, execute_script allows arbitrary JavaScript, and upload_file, cookie and navigation tools carry broad capabilities without documented sandboxing, confirmation, or data-flow boundaries, so least-privilege and confirmation are incomplete; score 8.
Maintenance 8/20
The repository is not archived and has an MIT license, with signs of ongoing maintenance in the README. However, no commit history, releases, or dependency updates are present in the supplied material, so maintenance activity cannot be confirmed and points are deducted; score 8.
Documentation 16/20
The README provides client-specific install examples, full tool tables with parameter types and defaults, resource descriptions, and development/test guidance, which is strong. It lacks troubleshooting, command limits, and security boundary documentation, and there is no verifiable source-level documentation, so a small deduction is made.
Setup experience 14/20
One-line npx installation and configuration examples for Goose, Claude Code, Cursor/Windsurf, and others make setup clear. Static review found no CI or committed tests as execution evidence, so per calibration setup cannot exceed 15; browser/driver prerequisites also remain, hence 14.

Static review · not runListed 2026-08-07

Read the FMRS scoring method →

Fit and risk

What it can accessWrites / deletes local filesRuns commands or codeUses the networkControls a browser

Best for

  • Developers and testers who want AI-assisted browser automation.
  • Scenarios requiring headless browsers for integration testing or crawling.
  • Users who prefer natural language commands over manual scripting.
  • Teams already using Selenium and looking to integrate AI agents.

Not for

  • Not suitable for scenarios requiring fine-grained control over browser internals (though execute_script exists).
  • Not suitable for tasks that don't need browser automation, like pure API testing.
  • Not ideal for production environments requiring official support or enterprise guarantees, as it's community-maintained.
  • Safari users need extra setup and lack headless mode, which may not fit all automation needs.

Required permissions

  • Network access to download browser drivers and communicate with browsers.
  • Local filesystem access for file uploads and screenshot saves.
  • Potential system permissions (e.g., `sudo safaridriver --enable`).
  • Execution permissions to run npx commands.

Risks and side effects

  • Browser automation may be detected by websites, leading to IP bans or ToS violations.
  • Headless mode can be recognized by some sites, affecting test validity.
  • The server itself is not malicious, but executing arbitrary JavaScript via execute_script poses security risks if the agent is instructed maliciously.
  • Dependency on npx may download unspecified versions, introducing supply chain risks.
  • Misuse could accidentally submit forms, delete data, or perform destructive actions.
  • Safari configuration requires admin privileges, which may modify system settings.

Setup

Before you start

Runtime：Node.js

Installation steps:

  1. Goose (Desktop): Paste in browser address bar: goose://extension?cmd=npx&arg=-y&arg=%40angiejones%2Fmcp-selenium%40latest&id=selenium-mcp&name=Selenium%20MCP&description=automates%20browser%20interactions.
  2. Goose (CLI): Run goose session --with-extension "npx -y @angiejones/mcp-selenium@latest".
  3. Claude Code: Run claude mcp add selenium -- npx -y @angiejones/mcp-selenium@latest.
  4. For Cursor, Windsurf, etc.: Add config {"mcpServers":{"selenium":{"command":"npx","args":["-y","@angiejones/mcp-selenium@latest"]}}}.
  5. Via Smithery: npx -y @smithery/cli install @angiejones/mcp-selenium --client claude.
  6. Global install: npm install -g @angiejones/mcp-selenium, then run mcp-selenium.
.mcp.json 
{"mcpServers":{"selenium":{"command":"npx","args":["-y","@angiejones/mcp-selenium@latest"]}}}

Shown for Claude Code. Other clients may use a different file or key (VS Code uses "servers") — the configurator below converts it.

.vscode/mcp.json
{
  "servers": {
    "selenium": {
      "command": "npx",
      "args": [
        "-y",
        "@angiejones/mcp-selenium@latest"
      ]
    }
  }
}

Goes in your project's .vscode/mcp.json (VS Code uses a "servers" key).

Terminal
claude mcp add selenium -- npx -y @angiejones/mcp-selenium@latest

Run it in a terminal; replace any <…> placeholders with your own values first.

Generate and validate client config

Check that it works

The client's tool list should show selenium's tools such as start_browser, navigate, and take_screenshot; or ask the AI to "open Chrome, go to github.com/angiejones, and take a screenshot" to confirm the connection works.

Troubleshooting

  1. Ensure Chrome or Firefox is installed and chromedriver is on PATH (for tests).
  2. If Safari fails to launch, run `sudo safaridriver --enable` and enable 'Allow Remote Automation' in Safari settings.
  3. Check that the npx command resolves the package; consider pinning a version.
  4. If browser fails to start, check firewall or proxy blocking.
  5. Review returned errors to verify parameters (locator strategy, timeouts).
  6. Check server logs in the MCP client for connection status.

Things to try

Once connected, you can ask your AI assistant things like:

  • Open Chrome, go to github.com/angiejones, and take a screenshot.
  • Start headless Firefox, navigate to example.com, and get the text of the main heading.
  • Type into the search box on the current page and press Enter.
  • List all cookies on this page, then delete the one named test.

Tools 18

start_browser writes
Launches a browser session.
navigate writes
Navigates to a URL.
interact writes
Performs a mouse action (click, double-click, right-click, hover) on an element.
send_keys writes
Types text into an element, clearing the field first.
get_element_text read-only
Gets the text content of an element.
get_element_attribute read-only
Gets an attribute value from an element.
press_key writes
Presses a keyboard key.
upload_file writes
Uploads a file via a file input element.
Show 10 more tools
take_screenshot writes
Captures a screenshot of the current page.
close_session writes
Closes the current browser session.
execute_script writes
Executes JavaScript in the browser for advanced interactions.
window writes
Manages browser windows and tabs.
frame writes
Switches focus to a frame or back to the main page.
alert writes
Handles browser alert, confirm, or prompt dialogs.
add_cookie writes
Adds a cookie.
get_cookies read-only
Gets cookies.
delete_cookie destructive
Deletes cookies.
diagnostics read-only
Gets browser diagnostics captured via WebDriver BiDi (console, errors, network).

Use cases

Ask an AI agent to open Chrome, navigate to a URL, and take a screenshot.
Automate data extraction from web pages, such as getting texts and attributes.
Run cross-browser tests to verify web app behavior.
Handle authenticated sessions and cookie management.
Run headless automation for continuous testing or scraping.

Supported clients

Goose
Claude Code
Cursor
Windsurf

Listed from the project's documentation, not tested by this site.

Overview

This is a Model Context Protocol (MCP) server that wraps Selenium WebDriver, enabling AI agents to control browsers through natural language instructions, performing navigation, clicks, input, screenshots, and more. It supports Chrome, Firefox, Edge, and Safari. No manual scripting is needed; agents can call the APIs to complete automation tasks.

Similar servers

Compare with other MCP servers

Source revision 363ecf0f7195 Data synced 2026-10-11 Read the FMRS scoring method