Tabby-MCP-Server

You set up Tabby-MCP-Server to let an AI assistant control and automate your terminal via the Model Context Protocol (MCP). It lets the AI read terminal output, manage SSH sessions, safely abort commands, and store command history, all with configurable safety and access controls.

How to use

Connect an AI client to your MCP server and start giving it tasks to manage your terminal sessions. You can ask the AI to list active sessions, run commands in a chosen session, and retrieve terminal output. If you enable Pair Programming Mode, the AI must prompt you for confirmation before executing commands and will auto-focus the terminal when commands run.

To begin using Tabby’s MCP server, you typically configure the AI client to connect to either the local stdio server or the SSE HTTP endpoint exposed by Tabby. Use the provided commands to start the local MCP server or connect to the SSE endpoint from your AI client.

How to install

Prerequisites: you need Node.js installed on your system.

Install from Tabby Plugin Store:

1. Go to Tabby settings → Plugins → MCP

2. Click "Install" on the Tabby MCP plugin

3. Restart Tabby

4. Configure your AI client to connect to the MCP server (see Connecting to MCP)

Using Docker to build and run the plugin locally:

git clone https://github.com/thuanpham582002/tabby-mcp-server.git
cd tabby-mcp-server
# Build the Docker image
docker build -t tabby-mcp . && docker run -v $(pwd)/build:/output tabby-mcp
bash scripts/copy_to_plugin_folder.sh

Additional configuration and notes

Configure the MCP server port and safety features in Tabby settings to tailor how the AI interacts with your terminal.

Pair Programming Mode adds safety features like a confirmation dialog before executing commands, autofocus behavior, and the ability to reject commands with feedback. Enable and adjust these options as needed in the MCP settings.

To connect an AI client, you can use either the local stdio server or the SSE (HTTP) endpoint. The local stdio server is started via a command shown in the setup examples, while the SSE endpoint is accessed over HTTP.

If you want the AI to control the terminal, configure your AI client to connect to the following MCP endpoints as shown in the examples.

# STDIO mode (local server via npx)
{
  "mcpServers": {
    "Tabby MCP": {
      "command": "npx",
      "args": [
        "-y",
        "tabby-mcp-stdio",
        "--port",
        "3001"
      ]
    }
  }
}

# SSE mode (HTTP endpoint)
{
  "mcpServers": {
    "Tabby MCP": {
      "type": "sse",
      "url": "http://localhost:3001/sse"
    }
  }
}

Available tools

get_ssh_session_list

Get a list of all terminal sessions that are currently active within the MCP-enabled environment.

exec_command

Execute a command in a specified terminal session and capture its output.

get_terminal_buffer

Retrieve content from the terminal buffer for a given session and range.

get_command_output

Retrieve the complete output of a previously executed command across pages or ranges.