SwarmTask is an MCP server that enables asynchronous parallel execution of shell commands through Model Context Protocol clients like Claude Code. Written in Go, it connects AI assistants to the local operating system by distributing command-line tasks across lightweight goroutines. Software engineers, DevOps professionals, and system administrators use SwarmTask to execute batch operations such as builds, test suites, multi-host network checks, and system diagnostics without blocking the assistant interface. Instead of waiting for long-running scripts sequentially, clients submit batches of up to fifty tasks at once and receive an immediate batch identifier. SwarmTask tracks each command in memory using thread-safe data structures and channels, letting AI models periodically poll execution status, review percentage progress, and inspect standard outputs or failure messages independently.
Category: Cloud & Infrastructure
Tags: automation, cli, parallel-processing, shell, task-management
bash ./swarmtask --port 3500 2. Open your Claude Code configuration file located at ~/.claude/mcp_servers.json. 3. Add the SwarmTask server configuration using supergateway: json { "mcpServers": { "swarmtask": { "command": "npx", "args": [ "-y", "supergateway", "--streamableHttp", "http://localhost:3500/mcp" ], "env": {} } } } 4. Restart Claude Code to load the tools.Part of MCP Servers
Run the compiled binary locally specifying the port with ./swarmtask --port 3500. Then, configure your client such as Claude Code by adding a supergateway entry targeting http://localhost:3500/mcp inside ~/.claude/mcp_servers.json and restart the client.
SwarmTask lets MCP clients dispatch batches of up to 50 shell commands asynchronously. It executes them concurrently in Go goroutines and provides tools to poll status, retrieve command output, and monitor batch progress percentages.
SwarmTask integrates with any client supporting Model Context Protocol streamable HTTP connections through supergateway, including Claude Code and Claude Desktop.
Commands are executed via the host shell with the exact same system privileges as the SwarmTask process. Script injection protection and sandboxing are not present, so commands run unrestricted on the local machine.