swarmtask

SwarmTask is an MCP server that enables parallel execution of shell commands via AI assistants. Through the MCP interface, you submit a batch of commands that are executed concurrently in Go goroutines, with real-time progress and per-task results. The server exposes two MCP tools: submit_tasks for launching multiple commands in parallel, and check_status for polling batch progress and retrieving results. Tasks are identified within a batch by a combination of the batch ID and the task name, and each task runs independently so a failure in one does not block others. This makes it ideal for orchestrating multiple system administration commands, development workflows, data processing steps, network checks, and monitoring tasks in one go. The server is designed to work with Claude Code via the supergateway for streaming results, but can be used by any MCP-enabled client that speaks the standard MCP tool format.

To use it, run the server (default port is 3500) and configure Claude Code (or your MCP client) to point to http://localhost:3500/mcp. You can then call submit_tasks with a JSON payload mapping task names to shell commands, and poll with check_status using the returned batch_id to monitor progress and fetch results in real time. The system supports up to 50 tasks per batch and executes all tasks in parallel, returning immediate batch IDs so you can begin monitoring while execution proceeds.

Prerequisites:

• A Unix-like environment with a Go toolchain available to build the binary, or use a prebuilt SwarmTask binary for your platform.
• Access to run executable files and bind to a port (default 3500).

Option A: Build from source

1. Install Go (https://golang.org/dl/).
2. Clone the repository: git clone https://github.com/rdhillbb/swarmtask.git cd swarmtask
3. Build the binary: go build -o swarmtask ./cmd/swarmtask
4. Run the server: ./swarmtask --port 3500

Option B: Use a prebuilt binary

1. Download the appropriate swarmtask binary for your OS from the releases page.
2. Make it executable: chmod +x swarmtask
3. Run the server: ./swarmtask --port 3500

Option C: Docker (if a image is provided)

1. Pull the image: docker pull yourorg/swarmtask:latest
2. Run the container (expose port 3500): docker run -p 3500:3500 yourorg/swarmtask:latest --port 3500

Notes and tips:

• Security: This release does not implement script injection protections or isolation, so commands run with the same privileges as the SwarmTask process. Exercise caution when executing untrusted commands.
• Environment variables: You can pass environment variables to the server process if needed. Some deployments may want to set STLOGDIR or other logging/env options via your runtime environment.
• Configuration: The MCP interface supports two tools (submit_tasks and check_status). Ensure your MCP client uses the correct batch_id and task_id naming conventions (batch_id-task_name).
• Performance: The server is designed for high concurrency with in-memory task tracking. For long-running or large workloads, monitor resource usage and consider upgrading CPU/RAM accordingly.
• Port configuration: If 3500 conflicts with other services, you can start SwarmTask on a different port with --port <port>.