Agent SSH MCP
MCP server for persistent SSH sessions with tmux integration, port forwarding, and SFTP file writes, built for AI agent workflows.
README
Agent SSH MCP
MCP server for persistent SSH sessions with tmux integration, port forwarding, and SFTP file writes — built for AI agent workflows.
Based on ssh-mcp-sessions by Justin Fry (MIT), with significant enhancements for agent reliability.
What's improved over the original
The original fryjustinc/ssh-mcp-sessions provides 8 basic tools that execute commands directly in a hidden ssh2 shell buffer. This repo is a significant rework:
Architecture changes
| Original | This repo |
|---|---|
| Commands run in a hidden ssh2 shell; output captured via in-memory buffer | Commands run inside a tmux session (ai); user can attach and watch live with tmux attach -t ai |
Single end marker: printf '__MCP_DONE__%s%d\n' $? — no start marker, no polling |
Dual start+end markers polled via capture-pane; output extracted between them |
| No reconnect logic — connection loss kills the session | Auto-reconnects on next command; keepalive every 30 s |
| No file transfer | SFTP: upload-file, download-file, write-remote-file |
| No port forwarding | forward-port / stop-forward via ssh2 forwardOut |
New tools added
| Tool | Purpose |
|---|---|
setup-tmux |
(Re)initialize tmux on an existing session |
upload-file |
Upload local file to remote via SFTP |
download-file |
Download remote file to local via SFTP |
write-remote-file |
Write text content directly via SFTP — no shell, no quoting issues, no heredoc |
forward-port |
Forward a local TCP port to a remote host:port |
stop-forward |
Stop a port forward tunnel |
Reliability fixes
| Issue | Root cause | Fix |
|---|---|---|
| Captured output contained entire pane history | End marker regex required a preceding \n; commands without trailing newline caused marker detection to fail → 30 s timeout → full pane returned as fallback |
Regex changed to `endMarker(\d+)(?:\n |
| Long sessions accumulated large scrollback, slowing polling | capture-pane -S -1000 scanned 1000 lines on every poll |
tmux clear-history before each command + window reduced to -S -200 |
heredoc (<< EOF) hung the session |
tmux waits for EOF input that never arrives | Detected and rejected with a helpful error |
exit code -1 was indistinguishable from a real command failure |
Timeout and true failure both returned the same code | Timeout now returns [tmux capture timed out — command may still be running] |
Installation
npm install -g agent-ssh-mcp
Or run without installing:
npx agent-ssh-mcp
MCP Client Configuration
Add to your MCP client config (Claude Desktop, Claude Code, Cursor, etc.):
{
"mcpServers": {
"ssh": {
"command": "npx",
"args": ["agent-ssh-mcp"]
}
}
}
Claude Desktop — ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows)
Claude Code / VS Code — .vscode/settings.json:
{
"claude.mcpServers": {
"ssh": { "command": "npx", "args": ["agent-ssh-mcp"] }
}
}
Cursor — ~/Library/Application Support/Cursor/mcp.json (macOS)
Host Configuration
Hosts are stored in ~/.ssh-mcp/hosts.json, created automatically on first use.
Add a host
{
"tool": "add-host",
"host_id": "my-server",
"host": "1.2.3.4",
"port": 22,
"username": "root",
"password": "optional",
"keyPath": "~/.ssh/id_rsa"
}
Authentication priority: password → private key → SSH agent (SSH_AUTH_SOCK).
Tools Reference
Session lifecycle
| Tool | Parameters | Description |
|---|---|---|
add-host |
host_id, host, port, username, password?, keyPath? |
Save SSH host config |
list-hosts |
— | List all saved hosts |
edit-host |
host_id + any field to change |
Update host config |
remove-host |
host_id |
Delete host config |
start-session |
host_id, sessionId? |
Connect and auto-initialize tmux |
list-sessions |
— | Show active sessions with uptime |
close-session |
sessionId |
Disconnect and clean up |
setup-tmux |
session_id |
(Re)initialize tmux on an existing session |
Command execution
| Tool | Parameters | Description |
|---|---|---|
exec |
session_id, command |
Run a command; auto-routes through tmux when available |
Note: Do not use tmux send-keys manually — exec handles tmux routing automatically.
Avoid heredoc syntax (<< EOF). Use write-remote-file instead.
File transfer
| Tool | Parameters | Description |
|---|---|---|
upload-file |
session_id, local_path, remote_path |
Upload local file via SFTP (preferred for large/binary files) |
download-file |
session_id, remote_path, local_path |
Download remote file via SFTP |
write-remote-file |
session_id, remote_path, content |
Write text content directly via SFTP — handles special characters, no shell quoting issues |
Port forwarding
| Tool | Parameters | Description |
|---|---|---|
forward-port |
session_id, local_port, remote_host, remote_port, tunnel_id? |
Forward 127.0.0.1:<local_port> → <remote_host>:<remote_port> |
stop-forward |
tunnel_id |
Stop a port forward tunnel |
Example — access a remote PostgreSQL locally:
{
"tool": "forward-port",
"session_id": "my-session",
"local_port": 15432,
"remote_host": "127.0.0.1",
"remote_port": 5432
}
Then connect to postgresql://user:pass@127.0.0.1:15432/db.
Typical Agent Workflow
1. add-host → save target server
2. start-session → SSH connect + tmux init (returns session_id)
3. exec → run commands (output is clean, no scrollback pollution)
4. write-remote-file / upload-file → write files reliably
5. forward-port → expose remote services locally if needed
6. close-session → clean up
Sessions auto-close after 2 hours of inactivity.
tmux Observability
Every command is executed inside a tmux session named ai. You (or the user) can attach to watch in real time:
ssh root@<host>
tmux attach -t ai
Building from Source
git clone https://github.com/xcdd/Agent-SSH-MCP
cd Agent-SSH-MCP
npm install
npm run build
Run tests:
npm test
Security Notes
~/.ssh-mcp/hosts.jsonmay contain passwords — treat it as sensitive (chmod 600).- Prefer key-based or SSH agent authentication.
- Sessions inherit all privileges of the SSH user.
License
MIT — see LICENSE.
Original work © 2025 Justin Fry. Modifications © 2026 xcdd.
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.