executor

Executor

Manage persistent interactive CLI processes with stateful stdin/stdout communication via MCP tools.

Pattern: Start → Send* → Read* → Stop

Prerequisites

This skill requires the executor-mcp MCP server to be installed and configured.

Installation:

pip install executor-mcp

Configuration: Add to Claude Desktop config (~/.config/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "executor": {
      "command": "executor-mcp"
    }
  }
}

When to Use

Scenario	Tool
REPL / debugger / database CLI / NuttX sim / QEMU	executor
One-off command that completes immediately	Bash
Need terminal emulation (curses, colors, vi)	tmux
Multiple concurrent interactive sessions to monitor	tmux

Use executor when:

• Process has a prompt and maintains state (Python REPL, gdb, psql, node)
• You need programmatic access to stdin/stdout without terminal complexity
• You want automatic buffering and logging of all I/O

Core Tools

All tools return JSON responses. Parameters shown below.

executor_start

Launch a new interactive process.

Parameters:

{
  "command": "python3",
  "args": ["-i"],
  "working_dir": "/optional/path"
}

Returns: process_id (unique identifier for subsequent operations)

executor_send

Send text to process stdin and optionally wait for output.

Parameters:

{
  "process_id": "abc123",
  "text": "x = 42",
  "wait_time": 0.1,
  "add_newline": true,
  "tail_lines": 20,
  "full_buffer": false
}

Key parameters:

• wait_time: Seconds to wait before reading (default: 0.1)
  • > 0: Wait and return NEW output from this command
  • 0: Send immediately, use executor_read_output later
• full_buffer: If true, return full buffer instead of just new output (default: false)
• tail_lines: Lines to return when using full_buffer=true (default: 20)

Returns: New output lines if wait_time > 0, otherwise "Success"