tabby

Tabby-MCP-Server implements a Model Context Protocol (MCP) server for Tabby Terminal, enabling an AI assistant to connect to your terminal, control sessions, and read outputs. The server exposes standard MCP interfaces such as listing SSH sessions, executing commands in a terminal, fetching terminal buffers, and retrieving complete command outputs. With the STDIO mode you can run the MCP server directly via npx, letting your AI client send and receive MCP messages through a local port (default 3001). In addition to the STDIO option, you can run the server via Docker or SSE if you prefer different connection methods. When connected, the AI client can request session lists, run commands, and access outputs in a structured, context-aware way, enabling safe and scripted automation of terminal tasks. The plugin also supports optional safety features like Pair Programming Mode to confirm potentially dangerous commands before execution and to focus the terminal after running actions.

Available tools exposed by the API include:

• get_ssh_session_list: Retrieve a list of all active terminal sessions.
• exec_command: Execute a given command in a specific terminal (tabId) with an optional explanation.
• get_terminal_buffer: Retrieve content from a terminal buffer by tabId and line range.
• get_command_output: Retrieve complete output for a previously executed command, with control over startLine and maxLines.

To connect from your AI client, configure an MCP connection pointing to the Tabby MCP server (default port 3001). If you run via the Tabby plugin, SSE connections are also supported, pointing to http://localhost:3001/sse. The recommended workflow is to start the server and then select the Tabby MCP (STDIO) option in your AI client, which will launch the server process on demand.

Prerequisites:\n- Node.js (v14+ recommended) and npm, or Docker if you prefer containerized usage.\n- Git for cloning the repository if you are building from source.\n\nInstall from npm (recommended for quick experiments):\n1) Ensure Node.js and npm are installed.\n2) Open a terminal and run:\n\nbash\nnpm install -g tabby-mcp\n 3) Start the server using STDIO mode (the default example in the README):\nbash\nnpx -y tabby-mcp-stdio --port 3001\n 4) Ensure the server is reachable on port 3001 and configure your AI client to connect to it.\n\nAlternative: Using Docker (build and run):\n1) Clone the repository and build the image:\nbash\ngit clone https://github.com/thuanpham582002/tabby-mcp-server.git\ncd tabby-mcp-server\ndocker build -t tabby-mcp .\n\n2) Run the container (adjust volumes as needed):\nbash\ndocker run -v $(pwd)/build:/output tabby-mcp\n 3) Copy built plugin files to the Tabby plugin folder if required (as described in the README).\n\nDevelopment workflow (optional):\n1) Clone the repo and install dependencies:\nbash\ngit clone https://github.com/thuanpham582002/tabby-mcp-server.git\ncd tabby-mcp-server\nnpm install\n\n2) Build and test with Docker as shown above.\n

Notes and tips:\n- The default STDIO server port is 3001; you can customize this via the --port flag. If you use SSE, ensure the server is running for the SSE URL http://localhost:3001/sse.\n- When using Tabby, you can enable Start on Boot to automatically start the MCP server when Tabby launches, and configure Pair Programming Mode for additional safety (confirm before executing commands).\n- The Tabby MCP README documents how to configure the client side mcp.json to connect via STDIO or SSE. If you run the Docker-based workflow, you can build the image and mount a build directory to extract the necessary plugin assets.\n- If you encounter port conflicts, choose an alternate port and update the client configuration accordingly.\n- Ensure the AI client supports MCP and is configured to use the same protocol version as the server.\n- For debugging, start the server with enableLogging set to true in Tabby settings to capture verbose activity.