OctSSH
Provides LLMs with safe, controllable, and stateful SSH access to shell environments, enabling remote command execution and file transfers with async support and security features.
README
OctSSH
I was just VibeCoding and suddenly thought: Why can't I just let my agent deploy code to the server for me?
So, I built OctSSH.
<img src="logo.png" width="200" />
OctSSH is an MCP server that gives LLMs safe, controllable, and stateful access to shell environments.
Note
By default (stdio mode), OctSSH only connects to machines already configured in your local ssh_config for passwordless login.
[!TIP] So... what makes OctSSH special?
Async Support
OctSSH provides a complete set of async tools to prevent LLMs from timing out on long-running tasks:
| Tool | Description |
|---|---|
exec(machine, command, confirm_code?) |
Run short commands synchronously |
sudo-exec(machine, command, confirm_code?) |
Run synchronously as root (sudo -n) |
exec-async(machine, command, confirm_code?) |
Run long tasks in background (screen) |
exec-async-sudo(...) |
Run background tasks as root |
write-stdin(session_id, data, append_newline?) |
Write to stdin of a running async task |
get-result(session_id, lines?) |
Inspect async task output |
grep-result(session_id, pattern, ...) |
Search task logs |
cancel(session_id) |
Terminate a task |
sleep(time) |
Pause (useful for polling) |
Note: In HTTP Serve mode, these tools operate directly on the local machine, and the
machineparameter is omitted.
Security Design
OctSSH features a Virtual Mode and Confirm Code verification flow:
π Safety Mechanism: Virtual Mode
We don't want AI to become a world-ending terminator, so we designed Virtual Mode.
When the AI attempts the following, OctSSH will not execute immediately, but instead returns a confirm_code:
- π File Overwrite: Uploading to a path that already exists.
- π High-Risk Commands:
rm -rfand similar "delete everything" commands. - π Regex Blocklist: Custom sensitive patterns defined in config.
Execution Flow Example:
- AI calls
exec("web", "rm -rf /var/www/html") - π OctSSH intercepts: Recursive delete detected -> Returns
confirm_code: a1b2c3+ file impact preview. - π€ User reviews and tells AI: "Confirm execution".
- β
AI calls
exec("web", "rm -rf /var/www/html", "a1b2c3")-> Actually executes.
Quick Start
Installation
npm install -g @aliyahzombie/octssh
octssh init
Usage Modes
1. Default Client Mode (stdio)
Runs locally and controls remote machines via SSH (reads ~/.ssh/config):
octssh
2. Streamable HTTP Server Mode (Local Control)
Install this on the target server. It exposes the server to LLMs via a secure HTTP interface. In this mode, OctSSH controls the local machine directly (no outbound SSH).
octssh serve
- Default Listen:
127.0.0.1:8787(Override viaOCTSSH_SERVE_HOST/OCTSSH_SERVE_PORT) - Auth: Prints a random key on startup. Clients must send header
X-OctSSH-Key: <key>.- Set fixed key:
export OCTSSH_SERVE_KEY="my-secret"
- Set fixed key:
- Tool Changes: Tools run on this machine.
machineparameter is omitted. SSH transfer tools (upload/download) are disabled.
Windows notes (serve mode)
On Windows, OctSSH serve will auto-select a local shell in this order:
sh(if available, e.g. Git-Bash/MSYS)pwshpowershellcmd
Override with:
set OCTSSH_SHELL=powershell
Tool Prefix (optional)
To avoid tool name collisions when you run multiple OctSSH instances, you can prefix all exposed tools:
export OCTSSH_TOOL_PREFIX="us1_"
Example: list becomes us1_list.
write-stdin (async interactive input)
write-stdin lets you send input to a running exec-async session.
Typical flow:
- Start a long-running command that reads stdin (via
exec-async) - Send data with
write-stdin(session_id, data) - Poll output with
get-result(session_id)
Notes:
- Default
append_newlineistrue. - Max payload is 64KiB per call.
- This is a streaming stdin: EOF is not sent. If the program exits on EOF, cancel the session instead.
- If you set
OCTSSH_TOOL_PREFIX, tool names are prefixed too (e.g.us1_write-stdin).
MCP Client Configuration
General (stdio)
Add to your MCP client config:
{
"mcpServers": {
"octssh": {
"command": "octssh",
"args": []
}
}
}
Claude Code CLI
claude mcp add octssh -- octssh
OpenCode CLI
{
"mcp": {
"octssh": {
"type": "local",
"command": "octssh",
"args": [],
"enabled": true
}
}
}
Or:
opencode mcp add octssh --command octssh
[!CAUTION] This project connects to real servers (or executes on the local machine). Please carefully review LLM operations. Using this project means you agree that the developer is not responsible for any accidental damage.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.