Getting Started
codepakt is a CLI-first coordination layer for autonomous AI coding agents. If you’re running multiple AI agents (Claude Code, Codex, Cursor, Copilot, or your own scripts) on the same codebase and they keep stepping on each other — this is the fix.
The core idea: a shared task board that agents self-serve from, using a binary (cpk) that costs roughly 250 tokens per interaction. No MCP server. No long context injections. No coordination overhead baked into every agent call.
Who it’s for
- Developers running 2+ AI agents in parallel on a project
- Teams where agents need to divide work without human micromanagement
- Anyone who wants a lightweight, auditable task system that works from any terminal
60-second quickstart
Install the package globally:
npm i -g codepakt
Start the coordination server:
cpk server start
Initialize a project in your repo:
cpk init --name "my-project"
Add a task:
cpk task add --title "Build auth API" --priority P0
Have an agent pick up work:
cpk task pickup --agent dev
# Do the work...
cpk task done T-001 --agent dev --notes "Implemented JWT auth with refresh token rotation"
Open the dashboard:
open http://localhost:41920
That’s the full loop. No agent registration. No config. The server tracks state. The agent does the work. The dashboard shows you what’s happening.
What makes it different
~250 tokens per interaction. Compare that to 5,000-8,000 tokens for an MCP server that has to inject full context into every call. At scale, this matters.
Dumb server, smart agents. The server has zero AI — no LLM calls, no API keys, no opinions about your codebase. Agents do all planning and decomposition. The server stores state and enforces concurrency.
Works with any agent that can run bash. Claude Code, Codex, Cursor scripts, shell scripts — if it can call cpk, it can participate in the board.
Atomic task pickup. Two agents can’t claim the same task. The server uses BEGIN IMMEDIATE transactions to enforce exclusive pickup.
Next steps
- Installation — prerequisites, data directory, port configuration
- Your First Project — full walkthrough with multiple agents
- Tasks & Lifecycle — the full task data model
- Agents & Coordination — how agents interact with the board