MCP Server

Skyvern MCP

Skyvern uses AI to navigate websites without explicit selectors. It 'understands' the page and decides what to click. Innovative but slower on complex multi-step flows.

Untested
View on GitHubby Skyvern

Install config

Claude Desktop (macOS)
{"mcpServers": {"skyvern": {"command": "npx", "args": ["-y", "@skyvern/mcp@latest"]}}}
Cursor
{"mcpServers": {"skyvern": {"command": "npx", "args": ["-y", "@skyvern/mcp@latest"]}}}
VS Code
// .vscode/mcp.json
{
  "servers": {
    "skyvern": {
      "command": "npx",
      "args": ["-y", "@skyvern/mcp@latest"]
    }
  }
}

Pros

  • +No selectors needed — AI-driven navigation
  • +Open source
  • +Innovative approach

Cons

  • Slow on complex DOMs
  • Lower reliability on untested sites
  • Higher token costs

Best for

Exploratory automation, rapidly changing UIs

Related servers in MCP Server

Frequently asked questions

How do I install Skyvern MCP?

Copy the config above into your AI assistant's MCP settings file:

  • Claude Desktop: ~/Library/Application Support/Claude/claude_desktop_config.json
  • Cursor: Settings → MCP → Add Server
  • VS Code: .vscode/mcp.json
Does it require an API key?

Yes. This server requires some form of configuration or credentials. Check the GitHub repository for setup instructions.

How were these results measured?

We run each server through a standardized test harness that:

  1. Installs the server via npx in a fresh environment
  2. Discovers available tools using the MCP SDK
  3. Executes real tasks (navigate, click, screenshot, etc.)
  4. Validates outputs with semantic assertions (not string matching)
  5. Measures latency percentiles and pass rate

All test scripts are open source on GitHub.

Related categories