home / mcp / agent communication mcp server

Agent Communication MCP Server

Provides a mission-control MCP server to coordinate multiple AI agents, enabling real-time progress tracking and transparent task execution.

Installation

Add the following to your MCP client configuration file.

Configuration

View docs

{
  "mcpServers": {
    "jerfowler-agent-comm-mcp-server": {
      "command": "npx",
      "args": [
        "@jerfowler/agent-comm-mcp-server"
      ],
      "env": {
        "AGENT_COMM_DIR": "./comm"
      }
    }
  }
}

You can coordinate multiple AI agents with this MCP server to delegate tasks, track progress in real time, and understand each step of the approach and execution. It acts as mission control, letting you assign specialized agents, observe their plans, and ensure work progresses smoothly without distracting your workflow.

How to use

Use an MCP client to delegate tasks to a configured agent and monitor progress through the task lifecycle. You will create tasks for specific agents, review their PLAN.md checklists, and observe PROGRESS updates as work advances. You can run multiple agents in parallel, and you can reconcile or adjust plans as needed to reach a successful outcome.

How to install

Prerequisites you need on your system are Node.js and npm (Node.js 14+ is typical, but check current recommendations for your environment). You may also use Python in some optional tooling contexts, but the MCP server itself operates with Node-based tooling in standard setups.

# Option 1: Global installation (recommended for MCP)
npm install -g @jerfowler/agent-comm-mcp-server

# Option 2: Use directly without installation
npx @jerfowler/agent-comm-mcp-server

# Option 3: From source
git clone https://github.com/jerfowler/agent-comm-mcp-server.git
cd agent-comm-mcp-server
npm install && npm run build

Setup with Claude (stdio MCP configuration)

To connect Claude Code to the MCP server you configure a stdio MCP server entry. The following example shows how to run the MCP server locally via npx and point Claude to use it.

{
  "type": "stdio",
  "name": "agent_comm",
  "command": "npx",
  "args": ["@jerfowler/agent-comm-mcp-server"],
  "env": {
    "AGENT_COMM_DIR": "./comm"
  }
}

Security and configuration notes

Do not commit API keys or credentials to any repository. The MCP configuration file can be gitignored to protect sensitive data. Use example templates for guidance and run the setup process to generate your configured file with secure defaults.

Try it out

After configuring your client, prompt Claude Code to delegate a task to a specific agent and begin monitoring progress. You can ask for full lifecycle visibility, including what the agent understood, how they planned, and the final result.

What to expect during use

You will see tasks move through INIT, PLAN, PROGRESS, and DONE/ERROR stages. You get automatic, non-blocking progress updates, and you can suspend or accelerate work as needed. The system emphasizes clean task context, transparency into agent thinking, and parallel execution where possible.

Available tools

mcp__agent_comm__submit_plan

Submit an implementation plan for a specific agent and validate it for timely execution.

mcp__agent_comm__track_task_progress

Track real-time progress for a given task and agent, returning status updates and percentages.

protocol_config

Configure protocol templates and task creation behavior to suit different kinds of work.

report_progress

Publish progress updates from an agent to keep the workflow visible and non-blocking.

sync_todo_checkboxes

Synchronize TodoWrite todo changes with PLAN.md checkboxes via the MCP server hook.

TaskArchive

Archive completed tasks and restore them if needed to keep the workspace clean.