Shell Use
Let your coding agent run shell commands in named tmux sessions the way you would—build, test, and debug without brittle one-off exec hacks.
Overview
shell-use is a MCP server for the Build phase that attaches agents to allowlisted tmux shell sessions over stdio so they can run human-like terminal workflows safely scoped by configuration.
What is this MCP server?
- stdio MCP server packaged as OCI image ghcr.io/kyuheon-kr/shell-use:1.0.0
- Session allowlist via required SHELL_USE_SESSIONS comma-separated names
- Optional SHELL_USE_SOCKET for tmux socket path (default /tmux/tmux.sock in Docker)
- Designed so agents drive shell workflows like a human operator
- Open-source implementation at github.com/kyuheon-kr/shell-use
- Required env SHELL_USE_SESSIONS (comma-separated session names)
- Server version 1.0.0 per registry metadata
What problem does it solve?
Coding agents struggle with real terminals—dev servers, interactive prompts, and session state—when you only give them one-shot command snippets.
Who is it for?
Solo builders who dockerize their dev environment and want Claude Code or Cursor to run long-lived shell workflows in named sessions.
Skip if: Beginners uncomfortable with shell safety, production servers without strict command policies, or workflows that need only read-only file tools.
What do I get? / Deliverables
After you configure SHELL_USE_SESSIONS and register the MCP server, agents can reuse persistent tmux sessions to build, test, and operate CLIs like you do locally.
- Agent-driven command execution inside named persistent shell sessions
- Configurable session boundaries via SHELL_USE_SESSIONS
- Docker-ready MCP integration for local build and test loops
Recommended MCP Servers
Journey fit
Interactive shell access matters while implementing and verifying features locally, before you automate everything in CI for ship. Agent-tooling captures MCP servers that extend what agents can operate—here, real terminal sessions instead of sandboxed snippet runners.
How it compares
Session-scoped terminal MCP, not a cloud CI runner or passive code-search skill.
Common Questions / FAQ
Who is Shell Use for?
Advanced solo builders and small teams who already use tmux or Docker dev images and want agents to execute allowlisted shell sessions during implementation.
When should I use Shell Use?
Use it while building and verifying features locally—running test suites, migrations, bundlers, or REPL-driven debugging—when persistent terminal state matters.
How do I add Shell Use to my agent?
Pull ghcr.io/kyuheon-kr/Shell Use:1.0.0, set required SHELL_USE_SESSIONS and optional SHELL_USE_SOCKET, register the stdio MCP entry in your agent config, and restart the client.