ssh-session-mcp bridges AI coding assistants with remote SSH terminals
Developer Zw-awa open-sourced `ssh-session-mcp`, an MCP server that gives AI coding assistants direct, shared access to persistent SSH PTY sessions on remote machines — eliminating the manual copy-paste relay between local AI tools and remote terminals.
Score breakdown
Developers using AI coding assistants on remote Linux machines, boards, or GPU servers can eliminate the manual copy-paste relay loop by letting the AI agent drive the SSH session directly through MCP tools.
- 01Zw-awa built and open-sourced `ssh-session-mcp`, an MCP server for managing persistent SSH PTY sessions accessible to AI agents.
- 02The core design uses one shared PTY so both the AI agent and the human developer see identical terminal state.
- 03A local browser-based terminal lets the developer observe and manually type into the same session without rebuilding it.
Zw-awa built `ssh-session-mcp` to solve a recurring friction point in AI-assisted development: the AI assistant lives locally, but the actual work runs on a remote machine. Without a bridge, the developer becomes a manual relay — switching to a terminal, SSHing in, running commands, and pasting output back into the AI chat. The project is an MCP server that manages a persistent SSH PTY session, exposing it to MCP clients so an AI agent can operate the remote shell directly through MCP tools like `ssh-quick-connect`, `ssh-run`, and `ssh-command-status`.
The core design decision is a single shared PTY rather than parallel lookalike sessions.
The core design decision is a single shared PTY rather than parallel lookalike sessions. Both the AI and the human developer observe and interact with the exact same terminal context via a local browser-based terminal. The tool accounts for real terminal complexity: it detects shell state (normal prompt, password prompt, pager, editor, long-running command), uses prompt detection combined with idle timeout and sentinel markers instead of a naive fixed timeout, and can automatically transition long-running commands to async mode for later polling. To prevent input collisions during human/AI handoffs, it supports switching between `common`, `user`, and `claude/codex` modes.
The project is deliberately narrow in scope, covering sessions, viewers, targeting, locks, logging, and tool contracts — and explicitly excluding domain-specific logic like ROS workflows, board-specific automation, training pipelines, or project orchestration. Zw-awa frames it as a small, reliable runtime layer installable via `npm install -g ssh-session-mcp`, most useful for embedded/board development, remote Linux development, GPU or VM-based workflows, and any AI-assisted setup where manual takeover of the remote terminal still matters.
Key facts
- 01Zw-awa built and open-sourced `ssh-session-mcp`, an MCP server for managing persistent SSH PTY sessions accessible to AI agents.
- 02The core design uses one shared PTY so both the AI agent and the human developer see identical terminal state.
- 03A local browser-based terminal lets the developer observe and manually type into the same session without rebuilding it.
- 04Supports `common`, `user`, and `claude/codex` modes to prevent input collisions during human/AI handoffs.
- 05Long-running commands can transition to async mode and be polled later via `ssh-command-status`.
- 06Command completion uses prompt detection, idle timeout, and sentinel markers rather than a fixed timeout.
- 07Installed via `npm install -g ssh-session-mcp`; scope is intentionally limited to transport/runtime, not domain-specific logic.
Topics
Summary and scoring are generated automatically from the original article. We always link back to the publisher and never republish images or paywalled content. Last processed Apr 21, 2026 · 18:16 UTC. How this works →