Skip to main content
Glama

🪖 ClaudeArmy

Version License Node

An MCP server that lets Claude orchestrate multiple Claude Code agents working autonomously across different projects.

ClaudeArmy Demo

What It Does

ClaudeArmy gives Claude (in the chat interface) the ability to spawn background Claude Code processes that work independently on coding tasks. Think of it as a chain of command:

  • You → give strategic direction

  • Claude (chat) → breaks it down and dispatches agents

  • Claude Code agents → execute autonomously, spawning their own sub-agents as needed

This means you can kick off work across multiple projects simultaneously, keep chatting normally, and check in on progress whenever you want.

Related MCP server: cc-agent

Tools

Tool

Description

dispatch_task

Deploy a Claude Code agent to a project directory. Supports mode: "team" for multi-agent coordination

check_tasks

Monitor status of all running/completed agents

get_task_output

Retrieve what an agent did and its full output

get_agent_team

Detailed view of a lead agent and all its sub-agents

cancel_task

Stop a running agent gracefully

purge_tasks

Clean up completed/failed tasks

Setup

  1. Clone this repo

  2. npm install

  3. Add to your Claude Desktop config:

    • Linux: ~/.config/Claude/claude_desktop_config.json

    • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

    • Windows: %APPDATA%\Claude\claude_desktop_config.json

"claude-army": {
  "command": "node",
  "args": ["/path/to/claude-army/src/index.js"]
}

Windows note: Use forward slashes in the path value (e.g. "C:/Users/you/claude-army/src/index.js").

  1. Restart Claude Desktop

Usage Examples

Single project task:

"Refactor GarbageFire's app.py into modular components"

Multi-project parallel work:

"Refactor GarbageFire to be more modular, and add a Salesforce adapter to my Reporting Tool"

Claude will dispatch separate agents to each project directory and let them work autonomously.

Check progress:

"How are the agents doing?"

Agent Teams

Use mode: "team" on dispatch_task to have the lead agent automatically decompose work and spawn specialized sub-agents that run in parallel.

How it works:

  • The lead agent analyzes the task and breaks it into focused subtasks

  • Each subtask is delegated to a sub-agent via Claude Code's Task tool

  • Sub-agents work in parallel, each with a well-defined scope

  • The lead agent synthesizes results once all sub-agents complete

Monitoring teams:

  • check_tasks shows a sub-agent summary (e.g. "3 sub-agents: 1 running, 2 completed")

  • get_agent_team gives a detailed tree view of the lead agent and every sub-agent, including status, runtime, and output previews

Configuration

Edit the constants at the top of src/index.js:

  • CLAUDE_BINARY - Path to your Claude Code binary

  • MAX_CONCURRENT_TASKS - Max simultaneous agents (default: 5)

Requirements

  • Claude Code installed and authenticated

  • Node.js 18+

  • Claude Desktop or any MCP-compatible client

Changelog

v0.4.1

  • Bump @modelcontextprotocol/sdk to ^1.26.0, resolving 3 high severity supply chain vulnerabilities

  • Add repository field to package.json

v0.4.0

  • Agent team support: track sub-agents spawned via Claude Code Task tool

  • New get_agent_team tool: tree view of lead + all sub-agents

  • check_tasks now shows sub-agent summary for team operations

  • dispatch_task gains mode parameter: solo (default) or team

  • Cross-platform: fixed Windows binary (claude.cmd) and signal handling

  • Error messages now show platform-appropriate diagnostic commands

  • README: added setup paths for Linux, macOS, and Windows

v0.3.0

  • Fixed stream-json parsing to correctly handle assistant message content arrays

  • Staleness indicator: check_tasks now shows time since last agent activity

  • Better error surfacing: actionable messages for CLI not found, permission denied, and mid-run failures

  • Fixed output duplication in get_task_output

  • Demo GIF added to README

v0.2.0

  • Real-time progress tracking via Claude Code's stream-json output

  • check_tasks now shows recent agent activity (file reads, edits, bash commands) instead of just runtime

  • get_task_output includes a full progress timeline with timestamps

  • Structured event parsing for tool use, assistant messages, and results

v0.1.0

  • Initial release

  • Core tools: dispatch_task, check_tasks, get_task_output, cancel_task, purge_tasks

  • Fire-and-forget architecture with background process management

  • Max 5 concurrent agents

  • stdio transport for MCP communication

License

MIT

Available Tools

6 tools
cancel_taskA

Cancel a running Claude Code agent task. Sends SIGTERM to gracefully stop the agent.

Inputs: task_id (required).

check_tasksA

Check the status of all deployed Claude Code agents. Shows which tasks are running, completed, or failed. Use this to monitor progress of background agents.

Inputs: status_filter.

dispatch_taskA

Deploy a Claude Code agent to work on a task in a specific project directory. The agent runs autonomously with full Claude Code capabilities including sub-agent spawning, file editing, bash execution, etc. Use this to delegate coding tasks to background agents. IMPORTANT: After dispatching, return to the conversation immediately. Do NOT call check_tasks or get_task_output unless the user explicitly asks for a status update.

Inputs: task (required), working_directory (required), mode, model, +1 more.

get_agent_teamA

Get a detailed view of a lead agent and all sub-agents it has spawned. Use this when a task is using Claude Code agent teams to see the full picture of what each sub-agent is doing, their status, and output previews. More detailed than check_tasks for multi-agent operations.

Inputs: task_id (required).

get_task_outputA

Retrieve the full output from a Claude Code agent task. Use this to see what the agent did, including its reasoning and actions taken.

Inputs: task_id (required), tail_lines.

purge_tasksA

Clear completed/failed tasks from the task list. Optionally clear all tasks (cancels running ones).

Inputs: include_running.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A4.1/5.0

    Scored across 6 tools

    Disambiguation4/5

    The tools are mostly distinct: dispatch_task creates, check_tasks monitors overall status, get_task_output retrieves output, cancel_task stops, purge_tasks cleans up, and get_agent_team drills into team hierarchy. However, check_tasks and get_agent_team both report status, with get_agent_team being more detailed, so there is minor potential for misselection.

    Naming Consistency5/5

    All tool names follow a consistent verb_noun pattern with underscores, such as check_tasks, get_task_output, and dispatch_task. The verbs are clear and related to task management, and the minor pluralization difference (task vs tasks) does not affect readability.

    Tool Count5/5

    Six tools is well within the ideal range for a focused task-management server. Each tool corresponds to a distinct action in the agent lifecycle—dispatch, monitor, retrieve, cancel, purge, and team inspection—so none feel redundant or missing.

    Completeness4/5

    The server covers the core lifecycle of Claude Code agents: creation (dispatch_task), status monitoring (check_tasks, get_agent_team), output retrieval (get_task_output), cancellation (cancel_task), and cleanup (purge_tasks). Minor gaps include no explicit retry or update mechanism, and get_agent_team only applies to team-based tasks, but the primary workflows are covered.

    Maintenance

    ActivityInactive
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers