Interactive Shell MCP
Enables LLMs to create and manage persistent, interactive shell sessions with full terminal emulation and PTY support. It allows for sequential command execution and supports interactive programs like vim or htop through specialized streaming and snapshot output modes.
README
Interactive Shell MCP
MCP server that provides interactive shell session management with full terminal emulation support via node-pty
Overview
The Interactive Shell MCP (Model Context Protocol) server enables LLMs to create and manage interactive shell sessions. It provides persistent shell environments where commands can be executed sequentially while maintaining state, similar to how a human would use a terminal.
Features
- Create and manage multiple concurrent shell sessions
- Full terminal emulation with proper TTY support
- Persistent shell state across commands
- Support for interactive programs (vim, nano, etc.)
- Cross-platform support (bash on Unix/Linux/macOS, PowerShell on Windows)
- Smart output handling with automatic mode detection
- Snapshot mode for continuously updating terminal applications
- Configurable output size limits to prevent memory overflow
- Automatic detection of terminal control sequences
Available Tools
start_shell_session
Spawns a new PTY shell and returns a unique session ID.
- Input: None
- Output:
{ sessionId: string }
send_shell_input
Writes input to the PTY with automatic newline handling.
- Input:
sessionId(string): The session ID of the shellinput(string): The input to send to the shell
- Output: Success confirmation
read_shell_output
Returns output from the PTY process with support for two modes:
-
Streaming mode (default): Returns buffered output since last read and clears the buffer
-
Snapshot mode: Returns the current terminal screen state without clearing (ideal for apps like top, htop, airodump-ng)
-
Input:
sessionId(string): The session ID of the shellmode(string, optional): Output mode - "streaming" (default) or "snapshot"maxBytes(number, optional): Maximum bytes to return (default: 100KB, max: 1MB)snapshotSize(number, optional): Size of the snapshot buffer to capture (default: 50KB)
-
Output:
{ "output": "string", "metadata": { "mode": "streaming|snapshot", "totalBytesReceived": number, "truncated": boolean, "originalSize": number, "isSnapshot": boolean, "snapshotTime": number } }
end_shell_session
Closes the PTY and cleans up resources.
- Input:
sessionId(string): The session ID of the shell to close
- Output: Success confirmation
Installation
npm install
npm run build
MCP Configuration
To use this MCP server with Claude Desktop or VS Code, add the following configuration to your MCP settings file:
Claude Desktop
Add to ~/Library/Application Support/Claude/claude_desktop_config.json on macOS or %APPDATA%\Claude\claude_desktop_config.json on Windows:
{
"mcpServers": {
"Interactive Shell MCP": {
"command": "node",
"args": [
"/path/to/interactive-shell-mcp/dist/server.js"
]
}
}
}
VS Code (Cursor)
Add to ~/.cursor/mcp.json:
{
"mcpServers": {
"Interactive Shell MCP": {
"command": "node",
"args": [
"/path/to/interactive-shell-mcp/dist/server.js"
]
}
}
}
Replace /path/to/interactive-shell-mcp with the actual path to your installation.
Usage Examples
Note: The examples below demonstrate how an LLM would interact with this MCP server. These are not JavaScript code to be run directly, but rather illustrate the expected tool calling patterns.
Working with High-Output Commands
When working with commands that produce large outputs or continuously refresh the screen (like airodump-ng, htop, top), use snapshot mode:
// Example of how an LLM would call these tools:
// Start a session
const { sessionId } = await start_shell_session();
// Run airodump-ng
await send_shell_input(sessionId, "sudo airodump-ng wlan0mon");
// Read output in snapshot mode to get current screen state
const result = await read_shell_output(sessionId, {
mode: "snapshot"
});
Handling Regular Commands
For normal commands that produce streaming output:
// Example of how an LLM would call these tools:
// Use default streaming mode
const output = await read_shell_output(sessionId);
// Or explicitly set a size limit for very large outputs
const output = await read_shell_output(sessionId, {
maxBytes: 50000 // Return only last 50KB
});
Output Modes Explained
- Streaming Mode: Best for regular commands. Returns all output since last read and clears the buffer.
- Snapshot Mode: Best for continuously updating applications. Returns the current terminal screen state without clearing. The server automatically switches to this mode when it detects terminal control sequences.
Debugging
To run the server independently for debugging:
npm start
This will start the server on stdio, which is primarily useful for testing the installation and debugging issues.
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。