Back-Agent

You can run development tasks through Claude Code CLI via the Back-Agent MCP Server. This server exposes a controlled, non-interactive task execution flow that you can trigger from an MCP client, with configurable working directories and timeouts, plus centralized logging and error handling to keep your workflow reliable.

How to use

You connect to the Back-Agent MCP Server from your MCP client to ask Claude Code CLI to perform development tasks. The server runs in non-interactive mode by default, so your client simply sends a task description and optional parameters, and the server handles execution, timeouts, and logging. You can customize the working directory for each task and set a timeout to ensure tasks do not hang.

To run a task, provide the task description and, if needed, a specific working directory and timeout. The Claude Code CLI will be invoked in a non-interactive mode automatically, so you do not need to supply a separate -p flag. The server will log activity and errors, giving you visibility into task success or failure.

How to install

Prerequisites you need before installing the MCP server are: Node.js version 18 or newer, and Claude Code CLI installed and available in your system path.

Step-by-step installation and setup commands you should run directly in your terminal.

npm install @zuens2020/back-agent-mcp

Additional setup and configuration

Configure how the server is launched from Claude Desktop by adding an MCP server entry in your Claude Desktop configuration. You can either run the server via Node with an explicit path to the built package, or run it through npx.

{
  "mcpServers": {
    "back-agent": {
      "command": "node",
      "args": ["--experimental-modules", "C:\\Users\\YourUsername\\AppData\\Roaming\\npm\\node_modules\\@zuens2020\\back-agent-mcp\\dist\\index.js"]
    }
  }
}

Alternative: using npx to run the MCP server

If you prefer using npx to run the server, add this configuration to Claude Desktop.

{
  "mcpServers": {
    "back-agent": {
      "command": "npx",
      "args": ["-y", "@zuens2020/back-agent-mcp"]
    }
  }
}

Notes on running in development or production

For development, you can run the server in development mode if you are testing changes. For production, build the project and run the built server. The server exposes a stable interface for MCP clients to request task execution.

Environment and logging

You can adjust logging verbosity using an environment variable. The supported value is LOG_LEVEL, with options like DEBUG, INFO, WARN, and ERROR.

Tools and endpoints

The server provides a single primary function to execute development tasks via Claude Code CLI. This is exposed as an MCP tool so MCP clients can describe the task and optional parameters, and receive the result or error.

Available tools

execute-task

Executes a development task using Claude Code CLI. You provide a task description, an optional working directory, a timeout, and any additional CLI arguments. The server handles non-interactive execution, timeout control, and logging.