ai-collab

You run an MCP server to enable direct AI-to-AI collaboration for project development. It coordinates autonomous agents, tracks tasks, plans, and looped execution so you can start a project and let agents handle the work with minimal human intervention.

How to use

You interact with the MCP server using an MCP client to start autonomous collaboration between AI agents. Start the agents with autonomous mode enabled to have them load a project context, generate a comprehensive plan, and begin executing tasks across multiple phases. You will monitor loop status to see progress and blockers, and you can intervene if needed.

How to install

Prerequisites: ensure you have Node.js installed on your system. You may also need a shell utility to run commands.

Step 1: Clone the project repository to your workspace.

Step 2: Install dependencies for the MCP server.

Step 3: Make the server executable if needed.

Configuration

Configure the MCP server for each AI client to enable local execution via Node. Below are explicit configuration examples you can adapt to your environment.

Claude configuration (local stdio server) builds the MCP command to run the server script from your project.

{
  "mcpServers": {
    "ai_collab": {
      "command": "node",
      "args": [".mcp-server/src/index.js"]
    }
  }
}

Security and best practices

Keep MCP server endpoints on trusted networks. Use appropriate access controls and audit logging for all autonomous actions.

Troubleshooting

If a command does not execute, verify the MCP client is pointing to the correct local stdio server entry and that the script path exists.

Project and data flow overview

The server coordinates a multi-phase project plan, handles task dependencies, prioritizes work, and provides continuous execution loops until completion or blockage. You can pause or adjust focus as urgent tasks arise.

Data storage

State and task data are stored under a structured data directory to persist across sessions. You can review task states, missions, and tickets as the project evolves.

Automation and usage patterns

Autonomous mode enables continuous work with minimal human input. You can start two AI agents in parallel and then periodically check their loop status to observe progress, blockers, and ticketing updates.

Supported tasks and commands

Use the following actions through your MCP client to manage work: creating tasks, batching tasks, reviewing submissions, starting project plans, and updating plan progress.

Example task flow

Create a batch of tasks with dependencies so that work can proceed in the correct order, and let the system automatically adjust when dependencies are satisfied.

Activate continuous work mode so the developer agent moves to the next available task immediately after completing one.

Autonomous loop status checks

Periodically query the agents to confirm they are progressing and to surface any critical blockers that require attention.

Final notes

This MCP server supports autonomous collaboration with AI agents, maintains context across sessions, and provides structured project planning and execution loops. Use it to accelerate development workflows with minimal manual intervention.

Available tools

send_directive

Create development tasks with optional dependencies and priorities to drive the project forward.

send_batch_directives

Queue multiple tasks at once to accelerate planning and execution.

review_work

Review submitted work and provide feedback or approval for progression.

create_project_plan

Instantiate a comprehensive project plan to guide autonomous execution across phases.

update_plan_progress

Advance the project to the next phase or adjust progress as tasks complete.

get_all_tasks

Query all tasks assigned to an agent, typically sorted by priority.

submit_work

Submit completed work items for validation and integration.

ask_question

Request clarification from agents to resolve ambiguities during execution.

get_loop_status

Check status of autonomous loops to observe progress and blockers.