Getting Started with UCP

Install once. Say "boot .ai". Never explain your project again.

The Problem You Know Too Well

Every time you start a new AI session, you're back to square one:

• "We're building a Next.js app with Supabase..."
• "The folder structure is frontend/, backend/, and docs/..."
• "Remember, we agreed to use Tailwind, not CSS modules..."

Sound familiar? You're not alone.

The average developer spends 15-20 minutes per session re-explaining context that the AI already knew yesterday. Multiply that by 10 sessions a week, and you're losing hours to repetition.

UCP fixes this.

What is UCP?

UCP (Unified Context Protocol) is a shared memory layer between you and your AI agent.

Think of it like this:

• Without UCP: Your AI has amnesia. Every session is a blank slate.
• With UCP: Your AI has a journal. It reads before starting, writes as it works.

The "journal" lives in a .ai/ folder in your repo. It's version-controlled, team-shareable, and works with any AI tool—Cursor, Copilot, Antigravity, Claude, and more.

How It Works

Step 1: Install UCP

Open your terminal in your project root:

npx corepackai install @corepackai/ucp

This creates a .ai/ directory with a pre-configured memory structure. You don't need to touch it—the agent handles everything.

Step 2: Tell Your Agent to Boot

In your next AI session (Cursor, Copilot, etc.), just say:

"boot .ai"

That's the magic phrase. Your agent will:

1. Read the boot protocol at .ai/boot.md
2. Load your project state from MASTER.md
3. Check your current priorities from PRIORITY.md
4. Review the changelog for recent work

Now it knows your project as well as you do.

Step 3: Work Normally

From here, just work. The agent:

• Remembers your tech stack
• Follows your coding standards
• Tracks what it changed in changelog.md
• Updates priorities as you complete tasks

No more re-explaining. No more "as I mentioned before..."

What's Inside .ai/?

After installation, your project gets this structure:

.ai/
├── boot.md              # Agent reads this first (don't edit)
├── context.json         # Registry of installed packs
├── modules/             # Pack source code (read-only)
└── context/             # Your project's memory
    └── corepackai/ucp/
        ├── MASTER.md    # Project overview & golden rules
        ├── PRIORITY.md  # Current focus & next steps
        ├── changelog.md # Auto-logged session history
        ├── tech.md      # Stack & dependencies
        └── active/      # In-progress work tracking

Key insight: You don't manually edit these files. The agent reads and writes to them during your sessions. It's a living document that grows with your project.

The Boot Flow

Here's exactly what happens when you say "boot .ai":

1. You say: "boot .ai"
2. Agent reads: .ai/boot.md
3. Agent finds: @corepackai/ucp installed
4. Agent loads: MASTER.md (project state)
5. Agent checks: PRIORITY.md (current focus)
6. Agent reviews: changelog.md (recent work)
7. Result: Agent is fully context-aware
8. You start collaborating — agent auto-logs to changelog.md

The entire boot process takes seconds. But it saves you hours.

Real Example

Before UCP:

"Hey, I'm working on a Next.js project with Supabase. We use TypeScript, Tailwind, and the app router. The main components are in frontend/components/. We have a blog section that uses markdown rendering. Oh, and we agreed to use the dark theme with #050505 as the base color..."

After UCP:

"boot .ai"

That's it. The agent already knows everything.

Verify Installation

Run this to confirm UCP is installed:

npx corepackai list

You should see:

📦 Installed Packages

  @corepackai/ucp @ 3.3.0
    Path: ./modules/corepackai/ucp

Quick Recap

No configuration. No manual file editing. Just seamless collaboration.

Want to Create Your Own Pack?

UCP is just one of many packs you can install. You can also:

• Fork UCP and customize it for your workflow
• Create your own packs with custom memory structures
• Publish to the Marketplace and share with the world

📚 Read the full guide: Creating & Publishing Packs — covering pack philosophy, pack.config.json, templates, and distribution.

For now, explore the source: 👉 github.com/breeznik/workflows/unified_context_protocol

Get Started Now

npx corepackai install @corepackai/ucp

Then tell your agent: "boot .ai"

Welcome to frictionless AI collaboration.