Control iTerm2 Terminals with AI

iTerm MCP Server lets an AI assistant create and manage iTerm2 terminal sessions directly: opening new terminals, executing commands in a specific terminal, reading back its output, listing all active terminals, and closing a terminal when done. All input is validated with Zod schemas, and commands run in isolated terminal sessions with sanitized input.

Use this connector when you want an assistant to run shell commands and read their output through real iTerm2 sessions - useful for interactive workflows where the assistant needs to keep a terminal open across multiple steps and inspect its live output.

Do not use it on non-macOS systems - iTerm2 is macOS-only, so this connector only works there.

open_terminal takes no required input and returns a terminal identifier. execute_command takes a terminal ID and a command string. read_output and close_terminal take a terminal ID. list_terminals takes no input. Outputs are terminal identifiers, command output text, or lists of active terminals with their info.

• open_terminal: open a new iTerm2 terminal instance
• execute_command: execute a command in a specific terminal
• read_output: read the output from a specific terminal
• list_terminals: list all active terminals and their information
• close_terminal: close a specific terminal

For Cursor, add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "terminal": {
      "command": "npx",
      "args": ["iterm_mcp_server"]
    }
  }
}

Requires Node.js 14+, macOS, and iTerm2 installed. No additional configuration is needed beyond standard MCP setup.

macOS developers who want an AI assistant to run and monitor shell commands in real iTerm2 terminal sessions.