HT MCP Server
Enables AI assistants to programmatically control interactive terminal applications through HT sessions, supporting session management, key sending, snapshots, and command execution.
README
HT MCP Server
A Model Context Protocol (MCP) server for interacting with HT (headless terminal) sessions. This allows AI assistants to programmatically control and interact with terminal applications through the HT interface.
What is HT?
HT is a project by Andy Konwinski and is the real magic behind this project. It wraps any binary with a VT100 style terminal interface and allows programmatic access via JSON commands. This MCP server provides a clean interface for AI assistants to create HT sessions, send commands, and capture terminal output.
Prerequisites
- HT (headless terminal) installed and available in PATH
- Node.js 18+
- TypeScript
Installation
# Clone and setup
git clone <this-repo>
cd ht-mcp-server
# Install dependencies
npm install
# Build the project
npm run build
Usage
Running the MCP Server
# Start the MCP server
npm start
# Or run in development mode
npm run dev
Available Tools
The server provides the following MCP tools:
ht_create_session
Create a new HT session.
- Arguments:
command(optional): Array of command arguments (default:["bash"])
- Returns: Session ID for use with other tools
ht_send_keys
Send keys to an HT session.
- Arguments:
sessionId: HT session IDkeys: Array of keys/text to send (supports special keys like "Enter", "Down", "^c", etc.)
ht_take_snapshot
Take a snapshot of the current terminal state.
- Arguments:
sessionId: HT session ID
- Returns: Current terminal content as text
ht_execute_command
Execute a command and return the output (convenience method).
- Arguments:
sessionId: HT session IDcommand: Command to execute
- Returns: Terminal output after command execution
ht_list_sessions
List all active HT sessions.
- Returns: List of active sessions with metadata
ht_close_session
Close an HT session and clean up resources.
- Arguments:
sessionId: HT session ID to close
Example Workflow
Here's how an AI assistant might use these tools to automate npm create vite:
// 1. Create a new HT session
const session = await ht_create_session({});
// 2. Navigate to a working directory
await ht_execute_command({
sessionId: session.sessionId,
command: "cd /tmp && mkdir vite-test && cd vite-test"
});
// 3. Start npm create vite
await ht_send_keys({
sessionId: session.sessionId,
keys: ["npm create vite@latest my-app"]
});
await ht_send_keys({
sessionId: session.sessionId,
keys: ["Enter"]
});
// 4. Wait and check what prompts appeared
const snapshot = await ht_take_snapshot({
sessionId: session.sessionId
});
// 5. Respond to prompts based on snapshot content
if (snapshot.snapshot.includes("framework")) {
await ht_send_keys({
sessionId: session.sessionId,
keys: ["Enter"] // Select default framework
});
}
// 6. Continue responding to prompts...
await ht_send_keys({
sessionId: session.sessionId,
keys: ["Down"] // Navigate to TypeScript variant
});
await ht_send_keys({
sessionId: session.sessionId,
keys: ["Enter"] // Select TypeScript
});
// 7. Verify the project was created
const result = await ht_execute_command({
sessionId: session.sessionId,
command: "ls -la"
});
// 8. Clean up
await ht_close_session({
sessionId: session.sessionId
});
Key Features
- Session Management: Create and manage multiple HT sessions
- Interactive Commands: Handle interactive CLI tools like
npm create,git, etc. - Terminal Snapshots: Capture terminal state to understand what's happening
- Special Key Support: Send arrow keys, Ctrl combinations, Enter, etc.
- Error Handling: Robust error handling and session cleanup
Integration with AI Assistants
This MCP server enables AI assistants to:
- Automate complex interactive CLI workflows
- Debug terminal applications by taking snapshots
- Handle multi-step processes that require user input
- Work with any CLI tool that has interactive prompts
Development
# Build
npm run build
# Development mode with hot reload
npm run dev
# Lint
npm run lint
# Clean build artifacts
npm run clean
Troubleshooting
- HT not found: Make sure HT is installed and in your PATH
- Session timeouts: Increase timeout values for slow commands
- Non-JSON output: The service filters out non-JSON lines automatically
- Process cleanup: Sessions are automatically cleaned up on server shutdown
License
ISC
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。
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。
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。
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。
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。
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。
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。
mcp-server-qdrant
这个仓库展示了如何为向量搜索引擎 Qdrant 创建一个 MCP (Managed Control Plane) 服务器的示例。