TermSSH MCP
Terminal-first SSH access for MCP clients and AI agents, enabling interactive remote sessions, file uploads, and stateful workflows.
README
<p align="center"> <img src="image/Banner.png" alt="TermSSH MCP Banner" width="100%" /> </p>
<h1 align="center">TermSSH MCP</h1>
<p align="center"> <strong>Terminal-first SSH access for MCP clients, AI agents, and remote automation.</strong> </p>
<p align="center"> Turn remote machines into agent-friendly, interactive terminal workflows — not just one-shot command execution. </p>
<p align="center"> <a href="https://github.com/rayss868/termssh-mcp"><img src="https://img.shields.io/badge/GitHub-rayss868%2Ftermssh--mcp-0d1117?style=for-the-badge&logo=github" alt="GitHub" /></a> <img src="https://img.shields.io/badge/MCP-Terminal--First-38bdf8?style=for-the-badge" alt="MCP Terminal First" /> <img src="https://img.shields.io/badge/SSH-Interactive-60a5fa?style=for-the-badge" alt="Interactive SSH" /> <img src="https://img.shields.io/badge/File%20Upload-Built--in-22c55e?style=for-the-badge" alt="File Upload" /> <a href="LICENSE"><img src="https://img.shields.io/badge/License-MIT-16a34a?style=for-the-badge" alt="MIT License" /></a> </p>
✨ Why TermSSH MCP
Most SSH tooling for AI workflows is built around run command → get output → done.
That falls apart when the real task is interactive:
- installers ask questions
- shells keep state
- debugging needs multiple steps
- deployments need uploads plus terminal control
- agents need to observe, react, and continue
TermSSH MCP is built for that gap.
Instead of pretending everything is a single command, it gives MCP clients a real operator-style workflow:
open a shell → write input → read output → keep context → upload files → continue working
🧠 What makes it different
<table> <tr> <td width="33%">
Terminal-first
Interactive terminal sessions are the core model, not an afterthought.
</td> <td width="33%">
Agent-ready
Designed for MCP clients, coding agents, and automation loops.
</td> <td width="33%">
Stateful workflows
Reuse active sessions so multi-step tasks feel natural and reliable.
</td> </tr> <tr> <td width="33%">
Upload included
Move scripts, configs, payloads, and generated artifacts over SFTP.
</td> <td width="33%">
Cross-platform
Works against Linux and Windows SSH targets.
</td> <td width="33%">
Clean tool surface
Focused MCP tools for terminal control and remote file delivery.
</td> </tr> </table>
🚀 Core capabilities
- Interactive SSH terminal sessions
- Incremental terminal read / write flow
- Managed terminal session reuse by default
- Optional forced multi-session creation
- Local file upload through SFTP
- Direct text and base64 content upload
- Terminal resize support
- Linux and Windows SSH target support
- MCP-native interface for AI tooling
🧰 Tool set
upload-file
Upload a local file from the MCP host machine to the remote SSH server using SFTP.
Parameters
localPath— local source file pathremotePath— destination path on the remote hostcreateDirectories— create missing parent directories if neededoverwrite— replace an existing remote file if presentmode— optional POSIX mode such as0644
upload-content
Upload direct text or base64 content to the remote server.
Parameters
content— raw text or base64 payloadencoding—utf8orbase64remotePath— destination path on the remote hostcreateDirectories— create missing parent directories if neededoverwrite— replace an existing remote file if presentmode— optional POSIX mode such as0644
terminal-start
Start an interactive remote terminal session.
Parameters
cwd— optional working directory after shell startupshell— optional shell binaryplatformHint—auto,linux, orwindowselevated— attemptsuelevation when configuredcols— terminal widthrows— terminal heightenv— optional environment variablesmultiSession— settrueto force a new managed session instead of reusing an existing one
terminal-write
Write input into an active terminal session.
Parameters
sessionId— target session idinput— text to sendappendNewline— append a newline automatically if needed
terminal-read
Read buffered output from a terminal session.
Parameters
sessionId— target session idsinceSequence— only return output newer than a given sequence numbermaxChars— limit the size of returned outputwaitForMs— optional short polling delay
terminal-resize
Resize an active terminal session.
Parameters
sessionId— target session idcols— new widthrows— new height
terminal-close
Close a terminal session locally.
Parameters
sessionId— target session id
🔄 Typical workflow
flowchart LR
A[terminal-start] --> B[terminal-write]
B --> C[terminal-read]
C --> D{Need file?}
D -- Yes --> E[upload-file / upload-content]
D -- No --> F{Continue session?}
E --> F
F -- Yes --> B
F -- No --> G[terminal-close]
This works especially well for:
- interactive package installs
- remote setup and provisioning
- deployments with artifact upload
- debugging services across multiple steps
- stateful shell workflows where context matters
🛠 Installation
Clone the repository
git clone https://github.com/rayss868/termssh-mcp.git
cd termssh-mcp
npm install
npm run build
Install globally
npm install -g termssh-mcp
⚙ Configuration
Required CLI parameters
host— hostname or IP address of the remote machineuser— SSH username
Optional CLI parameters
port— SSH port, default22password— SSH passwordkey— path to a private SSH keysudoPassword— optional password for sudo-oriented workflowssuPassword— optional password forsu-based elevationtimeout— SSH ready timeout in milliseconds, default60000maxChars— command-length validation limit, default1000; usenoneor0for unlimited mode
🧩 MCP configuration example
{
"mcpServers": {
"termssh-mcp": {
"command": "npx",
"args": [
"-y",
"termssh-mcp",
"--",
"--host=1.2.3.4",
"--port=22",
"--user=root",
"--password=pass",
"--timeout=30000",
"--maxChars=none"
]
}
}
}
SSH key example
{
"mcpServers": {
"termssh-mcp": {
"command": "npx",
"args": [
"-y",
"termssh-mcp",
"--",
"--host=example.com",
"--user=root",
"--key=/path/to/private/key"
]
}
}
}
🤖 Claude Code example
Register the server in Claude Code:
claude mcp add --transport stdio termssh-mcp -- npx -y termssh-mcp -- --host=YOUR_HOST --user=YOUR_USER --password=YOUR_PASSWORD
With SSH key authentication:
claude mcp add --transport stdio termssh-mcp -- npx -y termssh-mcp -- --host=example.com --user=root --key=/path/to/private/key
With extended timeout:
claude mcp add --transport stdio termssh-mcp -- npx -y termssh-mcp -- --host=192.168.1.100 --user=admin --password=your_password --timeout=120000 --maxChars=none
🎯 Great fit for
<table> <tr> <td width="33%">
Developers
- remote shell access from AI coding tools
- stateful debugging sessions
- script and config delivery
</td> <td width="33%">
DevOps / infra teams
- service inspection
- deployment support
- multi-step remote operations
</td> <td width="33%">
Agent builders
- terminal-native MCP workflows
- reusable sessions
- controlled remote automation loops
</td> </tr> </table>
🏗 Development
Build the project:
npm run build
Run tests:
npm test
Use the MCP Inspector:
npm run inspect
📁 Project structure
src/index.ts— MCP server entrypoint and tool registrationsrc/ssh-connection-manager.ts— SSH connection and terminal lifecycle handlingsrc/upload.ts— upload helpers and interactive session metadata helperssrc/core.ts— shared validation and SSH utility primitivestest/upload-and-terminal.test.ts— upload/session unit coveragetest/maxChars.test.ts— command validation coveragetest/smoke.ssh.test.ts— smoke tests for current exported behavior
🗺 Roadmap ideas
- richer session metadata inspection
- better remote session observability
- optional session persistence features
- more examples for Claude Code and MCP tools
- deployment-oriented workflow templates
🔐 Security note
TermSSH MCP gives remote access to systems over SSH.
Use it only on infrastructure you own or are explicitly authorized to manage.
📜 License
Released under the MIT License.
🤝 Contributing
Contributions are welcome.
See CONTRIBUTING.md for contribution guidance and CODE_OF_CONDUCT.md for expected behavior.
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。