Tabby MCP Server

A Tabby Terminal plugin that exposes your active terminal sessions through the Model Context Protocol (MCP). It lets MCP-compatible AI clients discover terminal tabs, execute commands, read terminal buffers, and retrieve long command output.

Tabby stays the terminal UI. Your AI client connects to Tabby through MCP.

Highlights

• MCP server inside Tabby — exposes an SSE endpoint at http://localhost:3001/sse
• Terminal session discovery — list local and SSH-backed Tabby terminal sessions
• Command execution — run commands in a selected Tabby terminal tab
• Robust output capture — simple marker protocol with exit-code parsing
• Terminal buffer access — read visible/history buffer ranges from a tab
• Long output pagination — retrieve full command output by outputId
• Pair programming mode — optional confirmation dialogs and user feedback
• Docker-based build — reproducible local plugin builds for Tabby

Demo

Table of Contents

• Requirements
• Installation
• Quick Start
• Connecting MCP Clients
  • SSE clients
  • Claude Code
  • Codex / OpenAI Codex CLI
• Tools
• Configuration
• Development
• License

Requirements

• Tabby Terminal installed and running
• Docker, for local builds
• macOS for the provided install script path
• An MCP-compatible client with SSE support, such as Claude Code, Cursor, Windsurf, Codex, or another SSE-capable MCP client

Installation

Option 1: Tabby Plugin Store

1. Open Tabby
2. Go to Settings → Plugins
3. Install Tabby MCP
4. Restart Tabby
5. Open Settings → Plugins → MCP and confirm the server configuration

Option 2: Local Docker Build

git clone https://github.com/thuanpham582002/tabby-mcp-server.git
cd tabby-mcp-server

make build-dist
bash scripts/copy_to_plugin_folder.sh

The install script copies the plugin to:

~/Library/Application Support/tabby/plugins/node_modules/tabby-mcp

Restart Tabby after installing or updating the plugin.

If your local Tabby installation requires bundled plugin dependencies, build the full package:

make build-dist-with-deps
bash scripts/copy_to_plugin_folder.sh

Quick Start

1. Start or restart Tabby
2. Open at least one terminal tab
3. Verify the MCP server is running:

curl http://localhost:3001/health
# OK

1. List Tabby terminal sessions:

curl -X POST http://localhost:3001/api/tool/get_ssh_session_list \
  -H 'Content-Type: application/json' \
  -d '{}'

1. Execute a command in a session:

curl -X POST http://localhost:3001/api/tool/exec_command \
  -H 'Content-Type: application/json' \
  -d '{"command":"pwd","tabId":"0"}'

Connecting MCP Clients

Tabby-MCP provides:

SSE endpoint:    http://localhost:3001/sse
Health endpoint: http://localhost:3001/health
HTTP test API:   http://localhost:3001/api/tool/<tool-name>

The Tabby plugin must be running before clients can connect.

SSE clients

Use this configuration for clients that support SSE MCP servers directly:

{
  "mcpServers": {
    "tabby-mcp": {
      "type": "sse",
      "url": "http://localhost:3001/sse"
    }
  }
}

For Cursor-style clients, place it in the MCP config file, for example:

~/.cursor/mcp.json

Claude Code

If your Claude Code version supports SSE MCP servers:

claude mcp add --transport sse tabby-mcp http://localhost:3001/sse

If your setup uses JSON config, use the SSE clients configuration.

Codex / OpenAI Codex CLI

Use the SSE MCP server configuration supported by your Codex client:

{
  "mcpServers": {
    "tabby-mcp": {
      "type": "sse",
      "url": "http://localhost:3001/sse"
    }
  }
}

For TOML-based configs, map the same SSE URL using your client's supported SSE MCP syntax.

Tools

Configuration

Default plugin configuration:

{
  "mcp": {
    "enabled": true,
    "startOnBoot": true,
    "port": 3001,
    "listenHost": "127.0.0.1",
    "serverUrl": "http://localhost:3001",
    "enableDebugLogging": true,
    "pairProgrammingMode": {
      "enabled": true,
      "showConfirmationDialog": true,
      "autoFocusTerminal": true
    }
  }
}

Configure these options in Tabby under Settings → Plugins → MCP.

To accept connections from other machines or environments such as WSL, set listenHost to 0.0.0.0 in the MCP settings. This makes the server reachable over the network; use an address reachable from your MCP client in its server URL and apply any needed firewall restrictions.

Pair Programming Mode

Pair Programming Mode adds safety prompts when an AI client executes commands:

• confirmation before command execution
• optional terminal auto-focus
• command rejection with feedback
• command result dialog and history tracking

Development

Clone and build:

git clone https://github.com/thuanpham582002/tabby-mcp-server.git
cd tabby-mcp-server
make build-dist

Install into local Tabby:

bash scripts/copy_to_plugin_folder.sh

Restart Tabby and verify:

curl http://localhost:3001/health

Build targets

make build-dist            # Build dist only
make build-dist-with-deps  # Build dist and copy node_modules
make help                  # Show available targets

License

MIT — see LICENSE.